No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Super-resolution land cover mapping using a Markov random field based approach
Abstract
Occurrence of mixed pixels in remote sensing images is a major problem particularly at coarse spatial resolutions. Therefore, sub-pixel classification is often preferred, where a pixel is resolved into various class components (also called class proportions or fractions). While, under most circumstances, land cover information in this form is more effective than crisp classification, sub-pixel classification fails to account for the spatial distribution of class proportions within the pixel. An alternative approach is to consider the spatial distribution of class proportions within and between pixels to perform super-resolution mapping (i.e. mapping land cover at a spatial resolution finer than the size of the pixel of the image). Markov random field (MRF) models are well suited to represent the spatial dependence within and between pixels. In this paper, an MRF model based approach is introduced to generate super-resolution land cover maps from remote sensing data. In the proposed MRF model based approach, the intensity values of pixels in a particular spatial structure (i.e., neighborhood) are allowed to have higher probability (i.e., weight) than others. Remote sensing images at two markedly different spatial resolutions, IKONOS MSS image at 4 in spatial resolution and Landsat ETM+ image at 30 in spatial resolution, are used to illustrate the effectiveness of the proposed MRF model based approach for super-resolution land cover mapping. The results show a significant increase in the accuracy of land cover maps at fine spatial resolution over that obtained from a recently proposed linear optimization approach suggested by Verhoeye and Wulf (2002) (Verhoeye, J., Wulf, R. D. (2002). Land Cover Mapping at Sub-pixel Scales using Linear Optimization Techniques, Remote Sensing of Environment, 79, 96-104). (c) 2005 Elsevier Inc. All rights reserved.
... SPM is a post-processing method applied to class fraction images and provides more information than spectral unmixing in land cover mapping. Various SPM methods have been proposed, including Markov random field (MRF) [43,44], Hopfield neural networks [45-47], pixel swapping algorithm (PSA) [42], spatial regularization [48], soft-then-hard model [49,50], attraction model [51,52], maximum a posteriori model [53], multi-objective optimization [44], spatial allocation [54], evolution algorithm [55], radial basis function interpolation [56], and deep learning [57,58]. Furthermore, SPM has been utilized in many fields, including mapping sub-pixel scale forests [59,60], trees [61], impervious surfaces [62], and surface water. ...
... Assuming the resulting sub-pixel map has the Markov random field property, the model assumes that neighboring pixels more possibly belong to the same land cover class than different classes. The MRF-SPM was first proposed by Kasetkasem et al. [44] and has been extended in many fields [36,41,62]. DE_MRF focuses on surface water mapping and adopts the water fraction images as the input, whereas the classic MRF-SPM in [44] focuses on multiple land covers and adopts the multispectral remote sensing images as the input. ...
... The MRF-SPM was first proposed by Kasetkasem et al. [44] and has been extended in many fields [36,41,62]. DE_MRF focuses on surface water mapping and adopts the water fraction images as the input, whereas the classic MRF-SPM in [44] focuses on multiple land covers and adopts the multispectral remote sensing images as the input. In particular, the MRF-SPM framework can be demonstrated in Equation (1): ...
Mapping high-spatial-resolution surface water bodies in urban and suburban areas is crucial in understanding the spatial distribution of surface water. Although Sentinel-2 images are popular in mapping water bodies, they are impacted by the mixed-pixel problem. Sub-pixel mapping can predict finer-spatial-resolution maps from the input remote sensing image and reduce the mixed-pixel problem to a great extent. This study proposes a sub-pixel surface water mapping method based on morphological dilation and erosion operations and the Markov random field (DE_MRF) to predict a 2 m resolution surface water map for heterogeneous regions from Sentinel-2 imagery. DE_MRF first segments the normalized difference water index image to extract water pixels and then detects the mixed pixels by using combined morphological dilation and erosion operations. For the mixed pixels, DE_MRF considers the intra-pixel spectral variability by extracting multiple water endmembers and multiple land endmembers within a local window to generate the water fraction images through spectral unmixing. DE_MRF was evaluated in the Jinshui Basin, China. The results suggested that DE_MRF generated a lower commission error rate for water pixels compared to the comparison methods. Because DE_MRF considers the intra-class spectral variabilities in the unmixing, it is better in mapping sub-pixel water distribution in heterogeneous regions where different water bodies with distinct spectral reflectance are present.
... However, this regularization optimization is an ill-posed problem. To convert the ill-posed image-based SRM to well-posed, examples include the Markov random field [37], fuzzy c-means [11,35], and spectral and spatial integration [36]. ...
... The performance of the proposed SGS is compared with that of HC and several state-of-art SRM algorithms, including the Markov random field-based (MRF) [37], spectral and spatial integration-based (SSI) [36], and CNN-based (CNN) [38]. MRF and SSI are implemented in MATLAB (version r2019b) (MathWorks. ...
... For benchmarks, HC was conducted by an SVM classifier, with samples chosen manually and randomly based on pixel numbers and spatial distributions from NLCD, containing a total of 10000 pixels. The parameters, settings, and selection of endmembers of MRF, SSI, and CNN were in accordance with previous studies [36][37][38]. MRF and SSI do not need training images; the CNN training dataset does not include land cover images and has only fine-resolution sub-pixel land cover maps and corresponding coarse-resolution remotely sensed imagery. ...
Super-resolution mapping (SRM) can effectively predict the spatial distribution of land cover classes within mixed pixels at a higher spatial resolution than the original remotely sensed imagery. The uncertainty of land cover fraction errors within mixed pixels is one of the most important factors affecting SRM accuracy. Studies have shown that SRM methods using deep learning techniques have significantly improved land cover mapping accuracy but have not coped well with spectral-spatial errors. This study proposes an end-to-end SRM model using a spectral-spatial generative adversarial network (SGS) with the direct input of multispectral remotely sensed imagery, which deals with spectral-spatial error. The proposed SGS comprises three parts: (1) Cube-based convolution for spectral unmixing is adopted to generate land cover fraction images. (2) A residual-in-residual dense block fully and jointly considers spectral and spatial information and reduces spectral errors. (3) A relativistic average GAN is designed as a backbone to further improve super-resolution performance and reduce spectral-spatial errors. SGS was tested in one synthetic and two realistic experiments with multi-/hyper-spectral remotely sensed imagery as the input, comparing the results with those of hard classification and several classic SRM methods. The results showed that SGS performed well at reducing land cover fraction errors, reconstructing spatial details, removing unpleasant and unrealistic land cover artifacts, and eliminating false recognition.
... Several SPM algorithms are designed to allow better estimates of the abundance and endmember, such as the Markov random field (MRF)-based SPM algorithm [26], the spectral-spatial integration SPM model [27], the adaptive MAP-based class This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ ...
... Assuming that the discrete class label field has MRF property, the land cover class occupying adjacent subpixels has a greater chance of belonging to the same land cover class membership [26]. In this sense, in order to encourage the fact that the subpixel l i,j and its neighborhood tend to have the same class label, the MRF-based multilevel logistic (MLL) prior can be adopted to model the spatial context information, which can be expressed as ...
... The proposed algorithm is evaluated on three HSIs, in comparison with several popular SPM methods, including the subpixel/pixel SAM [12], PSA [5], the spectral and spatial integration SPM model (SPMLM) [27],the MRF-based SPM model (MRFSPM) [26], as well as the genetic algorithm-based approach (GAAI) [29]. For all experimental data, we use the SCLSU to obtain the initial abundance maps for all SPM methods in this article. ...
Although Bayesian methods have been very effective for spatial–spectral analysis of hyperspectral imagery (HSI), they had not been fully explored for enhanced subpixel mapping (SPM) by simultaneously considering several key issues, i.e., endmember variability, the discrete nature of subpixel class labels, and the spatial information in HSI. Therefore, we propose a new Bayesian SPM method based on the discrete endmember variability mixture model (DEMM) and Markov random field (MRF), which has three main characteristics. First, DEMM allows the advanced SPM by completely accounting for the endmember–abundance patterns of each pixel to accommodate the endmember variability, the discrete hidden class label field of subpixels, while taking into account the noise heterogeneity effect. Second, the discrete class label fields modeled by MRF together with the DEMM, which can be integrated into a novel Bayesian model to better exploit the spatial contextual and spectral information. Third, the resulting Bayesian model can be efficiently solved by a designed expectation–maximization iteration, where E-step estimates the subpixel class label field using a simulated annealing algorithm and M-step estimates the endmembers for each pixel in HSI using the alternating non-negative least squares approach. The experimental results on three HSI datasets demonstrate that the proposed approach outperforms previously available SPM techniques.
... Following this principle, other regularization functions are introduced to handle the restoration step, such as Tikhonov-type regularizer [50], total variation (TV)-type regularizer [31,36,51], and nonlocal regularizers [32,42]. Also, the Markov random field (MRF) model, which includes the Gaussian MRF model and the Huber MRF regularizations, has been widely used in the context of super-resolution [1,7,18,25,26]. The Gaussian MRF model is considered a quadratic regularization term (similar to the Tikhonov model), which is known for its weakness related to the choice of a suitable threshold. Indeed, it is difficult to find an appropriate threshold for some real experiments. ...
... Figure 1 shows the visually most pleasant HR result when the optimal Lamé parameters λ and μ are presented with respect to their values of the function J (λ, μ) used to register the LR3 to LR1 image. As seen, the values J (λ, μ) reach their peaks when λ is considered in the interval [10,20] and μ is considered in the interval [25,30] simultaneously. This is confirmed by looking at the appearance of the function J with respect to the values of λ and μ. ...
The main idea of multi-frame super-resolution (SR) algorithms is to recover a single high-resolution image through a series of low-resolution ones of a captured scene. The success of the SR approaches is often related to well registration and restoration steps. In this work, we propose a new approach based on fluid optical flow image registration and a second-order regularization term to treat both the registration and restoration steps. The fluid registration is introduced to avoid misregistration errors, while the second-order regularization resolved by the Bregman iteration is employed to reduce the image artifacts. Moreover, we propose a bilevel supervised learning framework to compute the Lamé coefficients λ and μ, which perform the nonparametric registration of the super-resolution result. The numerical part demonstrated that the proposed method copes with some competitive SR methods.
... The Markov random field (MRF) is a technique that can be used to model contextual information). Kasetkasem et al [12] used the MRF to generate super-resolution for land cover mapping in remote sensing imagery. Every map is assumed to have Markov property. ...
... A minimum cut is a cut that has the minimum cost called min-cut and it can be achieved by finding the maximum flow which is verified in [12,13,14] that the min-cut is equivalent to max-flow. The max-flow/min-cut algorithm developed by Boykov and Kolmogorov [14] can be used to get the minimum cut for the s-t graph. ...
Remote sensing imagery classification has seen various problems such as the existence of mixed pixels. In fact, sub-pixel mapping proved recently to be an effective solution for this problem, whereas it provides a fine-resolution map of class labels from across spectrally unmixed fraction images. To improve sub-pixel mapping precision, we propose a method based on graph cuts.the problem can be transformed into a graph-based energy minimization problem and it can be solved using minimum cut/ maximum flow algorithm. Experimental results with one simulated data sets show the advantages of graph cuts.
... All these common degradations in HSIs limit the precision of the subsequent processing, such as classification, 1,2 unmixing, 3,4 subpixel mapping, [5][6][7] and target detection. 8,9 Compensating these degradations through quality improvement is therefore a key preprocessing step in the exploitation of HSIs. 10 With different degradation problems, radiometric quality improvement Thick cloud (f) Low spatial resolution methods for HSIs have been widely researched. ...
... The universal framework for fusion has been shown in Eqs. (5) and (6), which can be regarded as Bayesian framework and matrix factorization, respectively. For the Bayesian framework-based method in the original space, methods [106][107][108][109] use posterior distribution to estimate the required image according to the image prior. ...
... However, the algorithm does not fully consider the spatial dependencies under complex spatial semantics and the possible conceptual hierarchy of spatial attribute values, even if the neighboring relations are exploited in the classification. Existing research has explored two main approaches for incorporating spatial dependency into classification or prediction: (1) markov random field (MRF) model [14,17,43] and (2) spatial auto-regressive (SAR) model [2,20]. The MRF is a popular model for incorporating spatial context into image segmentation and land-use classification problems. ...
Traditional feature-based classification methods require objects to have the explicit, independent, and identifiable set of features, while most geo-referenced objects do not have the explicit features required by classifiers. Therefore, developing classificatory features under geospatial context is a prerequisite for effective spatial classification. Considering the spatial dependency, objects are correlated with each other, and for the object of interest its features (e.g., the distribution of neighboring objects) exist in a wide range of neighboring areas. However, the uncertainty of neighborhood size makes the dimensionality of potential feature set particularly high for spatial classification. Therefore, we propose a new model to automatically select a subset of spatially explicit features through continuous decision making by multiple agents in reinforcement learning (RL). A novel reward mechanism is developed to feed the knowledge of the downstream classification task back to the loop of feature selection. Through extensive experiments with facility points-of-interest datasets, we demonstrate that the subset of classificatory features selected by our RL model can help significantly improve the accuracy of spatial classification. Moreover, our feature selection has potential explainability for the spatial classification rules as it can determine the neighboring areas which have an impact on the classification result.
... Optimizer-based methods: These methods model the interactions between pixels in RGB images and depth maps represented by Markov random fields [12]. Optimization methods, such as the derived conditional random fields [39], total variation regularization [40], limited memory Broyden-Fletcher-Goldfarb-Shanno (L-BFGS) [41], approximate message passing [42], and non-negative matrix factorization [43], exhibit excellent performance in GDSR. However, these methods have the following limitations: high computational complexity, difficulty in adjusting parameters, and tendency to generate visual artifacts. ...
Depth images obtained from lightweight, real-time depth estimation models and consumer-oriented sensors typically have low-resolution issues. Traditional interpolation methods for depth image up-sampling result in a significant information loss, especially in edges with discontinuous depth variations (depth discontinuities). To address this issue, this paper proposes a semi-coupled deformable convolution network (SCD-Net) based on the idea of guided depth map super-resolution (GDSR). The method employs a semi-coupled feature extraction scheme to learn unique and similar features between RGB images and depth images. We utilize a Coordinate Attention (CA) to suppress redundant information in RGB features. Finally, a deformable convolutional module is employed to restore the original resolution of the depth image. The model is tested on NYUv2, Middlebury, Lu, and a Real-Sense real-world dataset created using an Intel Real-sense D455 structured-light camera. The super-resolution accuracy of SCD-Net at multiple scales is much higher than that of traditional methods and superior to recent state-of-the-art (SOTA) models, which demonstrates the effectiveness and flexibility of our model on GDSR tasks. In particular, our method further solves the problem of an RGB texture being over-transferred in GDSR tasks.
... One main category of the rst SRM group is based on Markov Random Filed (MRF) technique which was rstly developed in 2005 (Kasetkasem, Arora, and Varshney 2005). Regarding MRF concept, Kasetkasem et al. de ned an energy function conssists of two terms, entitled spectral energy and spatial enegrgy, and their contributions are controlled by a weight, or a smoothing parameter. ...
Super-resolution mapping (SRM) is a category of techniques that aim to estimate fine-scale land cover maps from coarse spatial resolution remote sensing images. The main limitations of SRM methods are high computational complexity, demanding training data and parameter tuning. To overcome these drawbacks, this paper proposes a cellular automata (CA) based SRM (SRM-CA) approach. CA is adopted as it is a fast and efficient technique that incorporates simple rules about spatial adjacency effects. In the first step of SRM-CA, the proportions of endmembers were computed, to generate SR map the pure pixels were then mapped. To assign an appropriate label for unlabeled sub-pixels; the energy function was computed. Each given sub-pixel was then assigned to a class with maximum amount of the energy. Two synthetic imageries, namely circle and concentric circles images, and an orthophoto map from the city Centre of Vaihingen, Germany were tested for validation and comparison. The average computed Percent Correct Classified (PCC’) index for Vaihingen dataset was 98.52%. Moreover, in the case of employed circle synthetic dataset, the comparison of the results between SRM-CA technique and SRM Using Neural Network Predicted Wavelet Coefficients model illustrated that there are no differences between PCC and Kappa coefficient. Regarding concentric circles, SRM-CA approach outperforms BPFM model with gains of 99.91% in Kappa metric. Meanwhile, the proposed method requires less than 50 seconds computation time for Vaihingen data set which considerably less than other state-of-the-art SRM methods, and hence SRM-CA approach provides a new solution to sub-pixel land cover mapping.
... The emergence of sub-pixel mapping (SPM) aims to generate a classification map by utilizing fractional abundance images to predict the locations of land cover classes within mixed pixels [18][19][20]. Numerous methods have been proposed, including those that employ structural similarity [21], pixel/sub-pixel spatial attraction model [22], pixel swapping algorithm [23], maximum a posteriori (MAP) [24,25] mode, Markov random field (MRF) [26,27], artificial neural network (ANN) [28][29][30], simulated annealing [31], total variant model [32], support vector regression [33], and collaborative representation [34]. However, it is important to note that these techniques only overcome spatial resolution limitations in specific applications. ...
... The most typical studies of this method are mathematical morphology [39], such as morphological transformations [40], morphological attribute profiles [41], extended morphological profiles (EMPs) [42], and multishape extended morphological profiles (MEMPs) [43]. In addition, Markov random field [44], [45], wavelet [46], [47], and level-set [48], [49] have been applied successfully in spatial feature exploration for HSRRS image classification. In these approaches, spatial information is assumed to meet the distribution of a mathematical model. ...
Spectral-spatial features are important for ground target identification and classification with High Spatial Resolution Remotely Sensed (HSRRS) Imagery. In this paper, two novel features, named the Gaussian-Weighting Spectral (GWS) feature and the Area Shape Index (ASI) feature, are proposed to complement the deficiency of the basic image feature for land cover classification with HSRRS imagery. The proposed GWS feature is an adaptive region-based feature that aims to improve the spectral homogeneity of a local area surrounding a pixel. Additionally, it is well known that the spectral feature is inadequate for classifying HSRRS imagery. Therefore, one spatial feature called the ASI feature is proposed here to describe the relationship between the area and shape for an adaptive region around each pixel. The proposed GWS and ASI features coupled with the basic red-green-blue feature are fed into a supervised classifier to obtain the final classification map. Experiments based on four real HSRRS images demonstrate that the proposed GWS and ASI features are capable of improving classification accuracies compared with some cognate state of the art methods. Moreover, the experiments also reveal that the proposed spectral-spatial features can complement each other for enhancing the classification performance with HSRRS images.
... These fraction images are further induced into a super-resolution map generator in order to get an initial resolution map at fine spatial resolution based on the following process. The pixels represented in coarse resolution image is in similar proportions to pixels corresponding to the finer resolution image 9,19,20 . If a fraction image of class A has a value of 0.5 in pixel sj, there are 0.5 a 2 pixels out of a 2 in the set j belonging to class A in the SRM. ...
... Furthermore, the optimization strategy SPM method takes the SPM problem as an optimization problem, iteratively generating the optimal SPM results using different objective functions and optimization strategies. Examples of optimization strategy SPM method include the SPM model based on Markov random field [25,26], the Hopfield neural network SPM model [27,28], SPM model based on maximum a posteriori (MAP) [29] and spectral-spatial constraint model [30]. These methods have achieved good results in alleviating the effect of abundance errors on the result of SPM. ...
For subpixel mapping (SPM), ensuring the operational efficiency of the algorithm and mitigating the effect of abundance errors often cannot be achieved simultaneously. To solve the problem, we propose a new SPM method based on the spatial adaptive attraction model (SAAM) and conditional random fields (CRFs). Firstly, the proposed SAAM obtains the spatial adaptive attraction value by adaptively adjusting the spatial attraction value obtained using the traditional spatial attraction model, thereby turning the display form of the abundance constraints in the SPM into an implicit form for expression, to perform the physical significance of the abundance constraints with the relative size of the attraction value of each subpixel. Secondly, the spatial adaptive attraction value of the implicitly represented abundance constraints and the local spatial smoothing prior are modelled in the CRFs, and the model makes full use of the spatial information in the label field while considering the abundance constraint. Thirdly, Graph-cut is used to optimize the model, make the proposed SPM can not only guarantee the operational efficiency, but also extinguish the influence of abundance error and decrease the noise artefact on the results of SPM. Experiments on three remote sensing images show that the proposed SPM accuracy is considerably better than the previously available SPM methods and is the least time-consuming. This study provides a new solution for the SPM of remote-sensing images.
... Generally, SRM relies on spatial prior knowledge to regularize the ill-posed problem of reconstructing LR to HR image (Atkinson 2009), which provides the indications of how each class located within a mixed pixel. We summarized four types of spatial prior: 1) Spatial dependence prior assumes that spatial proximate pixels more likely belong to the same class than the distant one, which includes spatial attraction model (Mertens et al. 2006), pixel swapping model (Atkinson 2005), genetic algorithm SRM (He et al. 2016a), Hopfield neural network SRM (Su 2019), spectral-spatial fusion-based SRM (Xu et al. 2018), and random field-based SRM (Kasetkasem, Arora, and Varshney 2005); 2) Geostatistical prior assumes that the sub-pixel distribution is subject to empirical geostatistical models that can be regressed with given samples (Boucher and Kyriakidis 2006), which includes multi-point simulation (Ge 2013), area-topoint kriging (Wang et al. 2015), and spatial distribution pattern-based SRM (Ge et al. 2016); 3) Regularization prior assumed that HR image follows the specific manifold like Laplacian model, total variation model, non-local prior or sparse prior, which includes maximum a posteriori SRM (Zhong et al. 2015), and sparse representation SRM ; 4) Spatiotemporal prior is aimed to provide complementary information from external dataset for constraint, such as time-series temporal images , shifted images (Zhong et al. 2014), panchromatic images (Nguyen, Atkinson, and Lewis 2011), and Point of Interest (POI) (Chen et al. 2018b). However, the spatial priors aforementioned are usually model-driven or expertly handcrafted, which is too inflexible for reconstructing the irregular heterogenous distribution (Ling et al. 2019b). ...
High resolution of global land cover dynamic is indicative for understanding the influence of anthropogenic activity on environmental change. However, most of the land cover products are based on Landsat image that only has 30 m resolution, which is insufficient to distinguish the heterogenous urban structure; while very high spatial resolution image usually has low temporal resolution, which is difficult to monitor the urban dynamic. Deep-learning-driven super-resolution mapping is a prevailing way of achieving very-high-resolution land cover dynamic products in aspect of alleviating the mixed pixel problem of Landsat image. However, two limitations are obvious: 1) the fixed grid of kernel during the upsampling process favors spatial homogeneity and suppresses the learning of spatial heterogeneity of urban composition and 2) geometric or radiation variation over large spatial and long temporal extent in remote sensing images makes the super-resolution mapping approach difficult to transfer for application. Here, we attempt to solve these two limitations: 1) a progressive edge-guided super-resolution architecture is designed to allow nonuniformed kernel specific at the low-confidence edge region and intensify the learning of heterogenous compositions’ patterns and 2) an alternating optimization strategy is designed to minimize the resultant entropy and modulate the classification hyperplane to accommodate to the manifold of the discrepant region. Validation experiments are investigated based on a fine-grained and large-extent super-resolution (FLAS) dataset constructed in this study, and it is found that our approach remarkably enhances rich detailed patterns of heterogenous region and outperforms other state-of-the-art algorithms. Besides, we applied DETNet to the large spatial extent of 28 metropolises in China (>40,000 km²) and the large temporal extent of continuous 21-year (2000–2020) in Wuhan city to examine transferability. From the land cover areas variation, we find that the expansion rate of cropland is faster than the urban expansion over the past 10 years, which are gradually becoming the principal source for the encroachment of forest and lakes. From detailed urban dynamic reflected by the 21-year products, we find that urban-villages between the old city zone and the outer high-tech development zone are gradually disappeared. The captured dynamic is consistence with the urban-village renovation policy during this period, which is meant to redistribute the spatial configuration of the city for a more sustainable urban structure. We believe that the proposed method can facilitate a seamless and fine-grained observation system that can fill the weakness of the existing land cover activities and provide a brand-new insight into the urban dynamic and its underlying mechanism.
... However, the subpixel mapping based on MRF considers both the spatial constraints of land cover classes and the spectral constraints of mixed pixels to directly obtain the subpixel mapping result and is independent of the low-resolution abundance image after spectral unmixing [76]. Tolpekin et al. conducted a more in-depth study on this model, analyzed the uncertainty of subpixel mapping results based on the MRF model [77], discussed the infuence of different spatial constraint and spectral constraint parameters on the mapping results [78], and studied the infuence of spectral separability of land cover classes on mapping results [79]. At the same time, this model is successfully applied to the extraction of tree information from RS images [80]. ...
... e Markov Random Field (MRF) [7][8][9] and the Fractal Dimension [10][11][12] are two approaches to texture segmentation that have stood the test of time. Because the MRF technique considers the texture picture to be a two-dimensional random process and assumes that each pixel's gray-value is solely related to that of the surrounding pixels, the texture image is modeled as a two-dimensional MRF model [13,14]. is is because the MRF technique assumes that each pixel's gray value is solely related to that of the surrounding pixels. ...
Traditional texture cluster algorithms are frequently used in engineering; however, despite their widespread application, these algorithms continue to suffer from drawbacks including excessive complexity and limited universality. This study will focus primarily on the analysis of the performance of a number of different texture clustering algorithms. In addition, the performance of traditional texture classification algorithms will be compared in terms of image size, clustering number, running time, and accuracy. Finally, the performance boundaries of various algorithms will be determined in order to determine where future improvements to these algorithms should be concentrated. In the experiment, some traditional clustering algorithms are used as comparative tools for performance analysis. The qualitative and quantitative data both show that there is a significant difference in performance between the different algorithms. It is only possible to achieve better performance by selecting the appropriate algorithm based on the characteristics of the texture image.
... However, the sub-pixel land cover map can be produced directly from the raw MS image. Kasetkasem et al. (2003Kasetkasem et al. ( , 2005 used the MS image directly as input for an optimisation process based on Markov random fields and a simulated annealing algorithm to produce the sub-pixel land cover image. ...
p>New approaches for using supplementary data such as panchromatic and fused imagery and Light Detection And Ranging (LiDAR) elevation data to increase the accuracy and spatial resolution of the thematic map were developed in this thesis. Information from the fused and panchromatic imagery was incorporated into the Hopfield neural network (HNN) model based forward and inverse models in form of reflectance functions. For the fused image, the forward and inverse models were formulated based on a linear mixture model and local end-member values. The reflectance function for the panchromatic image was derived locally based on the spectral and spatial convolutions. Visual and statistical analyses demonstrated that the use of fused and panchromatic imagery can increase accuracy of the sub-pixel image. The HNN super-resolution mapping using LiDAR elevation data is based on an optimisation process with a probability maximisation for the building class as a goal together with the goal functions and constraints of the traditional super-resolution mapping. The results showed a considerable increase in all accuracy statistics of the new technique, particularly for building objects.
Adopting the HNN model and forward model mechanism, three approaches for super-resolving of fine sub-pixel multispectral (MS) image from the coarse MS imagery were developed based on the HNN super-resolution mapping technique with the forward model and semivariogram matching. The first approach can be applied to predict the sub-pixel image based on the super-resolution of the mixed pixels. The second approach can be used to create the sub-pixel MS image with spectral features of the coarse resolution image and spatial variation at sub-pixel resolution. The aim of the third approach is to generate a smoothed sub-pixel image by maximising the spatial dependence between the sub-pixels on a semivariance value of zero at lag h =1.</p
... Super-resolution mapping (SRM) is a promising way to obtain high-spatial resolution land cover maps from coarse remotely sensed images [31]. In SRM, a mixed pixel is divided into a number of subpixels based on a zoom factor. ...
Water body mapping is an effective way to monitor dynamic changes in surface water, which is of great significance for water resource management. Super-resolution mapping is a valid method to generate high-resolution dynamic water body maps from low-spatial-resolution images. However, the accuracy of existing super-resolution mapping methods is not high due to the low accuracy of fraction images and the insufficiency of spatial pattern information. To solve this problem, this paper proposes a spectral similarity scale-based multiple-endmember spectral mixture analysis (SSS-based MESMA) and a multiscale spatio-temporal dependence method based on super-resolution mapping (MESMA_MST_SRM) for water bodies. SSS-based MESMA allows different coarse pixels to have different endmember combinations, which can effectively improve the accuracy of spectral unmixing and then improve the accuracy of fraction images. Multiscale spatio-temporal dependence adopts both pixel-based and subpixel-based spatial dependence. In this study, eight different types of water body mappings derived from the Landsat 8 Operational Land Imager (OLI) and Google Earth images were employed to test the performance of the MESMA_MST_SRM method. The results of the eight experiments showed that compared with the other four tested methods, the overall accuracy (OA) value, as well as the overall distribution and detailed information of the water map generated by the MESMA_MST_SRM method, were the best, indicating the great potential and efficiency of the proposed method in water body mapping.
... At the beginning of the century, scaling issues became more prominent, and several groups advanced the field in the search to address the challenging issue of downscaling spatial continua (Cressie, 1996;Kyriakidis, 2004;Pardo-Igúzquiza et al., 2006;Goovaerts, 2006Goovaerts, , 2007Atkinson et al., 2008;Atkinson, 2013;Hutengs and Vohland, 2016;Wang et al., 2016a,b;Jeganathan and Mondal, 2017). Even greater effort was paid to the challenge of downscaling reflectance to categories (referred to as sub-pixel mapping) (Atkinson, 1997;Tatem et al., 2001Tatem et al., , 2002Atkinson, 2005;Khasetkasem et al., 2005;Thornton et al., 2007;Tolpekin and Stein, 2009;Ardila et al., 2011;Nguyen et al., 2011;Su et al., 2012;Ling et al., 2013;Ai et al., 2014;Wang et al., 2014;Hu et al., 2015;Ge et al., 2016;Chen et al., 2018). Both change of support goals aim to escape the strictures of the pixel in remote sensing. ...
This paper summarizes the development and application of spatial statistical models in satellite optical remote sensing. The paper focuses on the development of a conceptual model that includes the measurement and sampling processes inherent in remote sensing. We organized this paper into five main sections: introducing the basis of remote sensing, including measurement and sampling; spatial variation, including variation through the object-based data model; advances in spatial statistical modelling; machine learning and explainable AI; a hierarchical ontological model of the nature of remotely sensed scenes. The paper finishes with a summary. We conclude that optical remote sensing provides an important source of data and information for the development of spatial statistical techniques that, in turn, serve as powerful tools to obtain important information from the images.
See the website: https://www.sciencedirect.com/science/article/pii/S2211675322000306?dgcid=coauthor
... (Kasetkasem et al., 2005) آﻣــﺎر زﻣــﻴﻦ و (Boucher & Kyriakidis, 2006) ﻧﻤـﻮد (Banavar et al., 2001) . Yatsu, 1995); ( Bui and Moon, 1995) . ...
... V. Potapov et al., 2012;Spracklen & Spracklen, 2021). To improve accuracy of land cover mapping, methods including multi-source data fusion, ensemble learning, object-based classification, postprocessing, and transfer learning were exploited by previous studies (Dorren, Maier, & Seijmonsbergen, 2003;Kasetkasem, Arora, & Varshney, 2005;Knauer et al., 2019;X. Li, Du, & Ling, 2013;Qin et al., 2015;Solberg, Taxt, & Jain, 1996;Van Coillie, Verbeke, & De Wulf, 2007;Zhang et al., 2017). ...
Global forest cover is critical to the provision of certain ecosystem services. With the advent of the google earth engine cloud platform, fine resolution global land cover mapping task could be accomplished in a matter of days instead of years. The amount of global forest cover (GFC) products has been steadily increasing in the last decades. However, it's hard for users to select suitable one due to great differences between these products, and the accuracy of these GFC products has not been verified on global scale. To provide guidelines for users and producers, it is urgent to produce a validation sample set at the global level. However, this labeling task is time and labor consuming, which has been the main obstacle to the progress of global land cover mapping. In this research, a labor-efficient semi-automatic framework is introduced to build a biggest ever Forest Sample Set (FSS) contained 395280 scattered samples categorized as forest, shrubland, grassland, impervious surface, etc. On the other hand, to provide guidelines for the users, we comprehensively validated the local and global mapping accuracy of all existing 30m GFC products, and analyzed and mapped the agreement of them. Moreover, to provide guidelines for the producers, optimal sampling strategy was proposed to improve the global forest classification. Furthermore, a new global forest cover named GlobeForest2020 has been generated, which proved to improve the previous highest state-of-the-art accuracies (obtained by Gong et al., 2017) by 2.77% in uncertain grids and by 1.11% in certain grids.
... Pan-sharpening [11] and super-resolution [12]). It can include filtering approaches such as fusing information within a local window using methods such as interpolation [13,14], maximum a posteriori (MAP), Bayesian model, Markov random fields (MRFs), and Neural Networks (NN) [4,12,[15][16][17][18]. Although spatialspectral fusion is efficient, it is not able to incorporate information from temporal images which produce dramatic radiometric differences such as those introduced by meteorological, phenological, or ecological changes. ...
Remote sensing images and techniques are powerful tools to investigate earth surface. Data quality is the key to enhance remote sensing applications and obtaining a clear and noise-free set of data is very difficult in most situations due to the varying acquisition (e.g., atmosphere and season), sensor, and platform (e.g., satellite angles and sensor characteristics) conditions. With the increasing development of satellites, nowadays Terabytes of remote sensing images can be acquired every day. Therefore, information and data fusion can be particularly important in the remote sensing community. The fusion integrates data from various sources acquired asynchronously for information extraction, analysis, and quality improvement. In this chapter, we aim to discuss the theory of spatiotemporal fusion by investigating previous works, in addition to describing the basic concepts and some of its applications by summarizing our prior and ongoing works.
... The demand of utilization of remote sensing data to observe the all type changes have increased over recent years [17]. Management of actual and regularly land use cover and change deduction information is necessary for natural resource management monitoring and planning program [18]. ...
A Hybrid approach of Land-use/land-cover (LU/LC) classification has been carried out in the study using Resource Sat-2 LISS-III Satellite data having a resolution of 23.5 m. With the help of ERDAS Imagine image processing software, unsupervised classification process of the satellite image has been made to obtain the required land use/land cover classes. During the classification study, a common mixed class assigned as Common Observed Classes (COC) used for the separation of the required classes to obtain undefined mixed land use/land cover classes with the help of ArcGIS model plate tool. Furthermore, more accuracy gained in the classification of separated mixed land cover classes by revising unsupervised classification processes. After that pre-defined unsupervised land cover classes and COC unsupervised classes were merged by using ArcGIS model plate tool to generate a new classified image with enhanced accuracy. Finally, the comparison between both supervised and COC hybrid unsupervised classes was made and better results are seen in hybrid unsupervised classification. Therefore, application of this new approach in land use/land cover studies, ground cover mapping, forest management, water resources management, sustainable urban development, agricultural studies, natural resources management and manmade resources development management planning change deduction etc. will provide better results as compared to other classification approaches techniques.
(11) (PDF) Geoinformatics & Geostatistics: An Overview Development of Hybrid Unsupervised Classification Techniques for Accuracy enhancement of Land Use/ Land Cover Mapping Using Geo-spatial Technology: Hoshangabad, District Madhya Pradesh India. Available from: https://www.researchgate.net/publication/350192014_Geoinformatics_Geostatistics_An_Overview_Development_of_Hybrid_Unsupervised_Classification_Techniques_for_Accuracy_enhancement_of_Land_Use_Land_Cover_Mapping_Using_Geo-spatial_Technology_Hoshangabad_D [accessed Mar 31 2021].
... (Kasetkasem et al., 2005) آﻣــﺎر زﻣــﻴﻦ و (Boucher & Kyriakidis, 2006) ﻧﻤـﻮد (Banavar et al., 2001) . Yatsu, 1995); ( Bui and Moon, 1995) . ...
... Subpixel mapping has been successfully applied in multiple remote sensing image applications [12], [13]. Kasetkasem et al. used Markov random fields (MRFs) to describe the statistical correlation between pixels by considering spatial and spectral constraints and realized subpixel mapping by calculating the probability of the various types of features belonging to subpixels [14]. Verhoeye et al. studied the mathematical model of the spatial correlation theory and transformed subpixel mapping into a linear optimization problem. ...
The fine extraction of water boundaries is of great significance for water resource monitoring, water environment monitoring, and flood prevention. MODIS images are widely used for water extraction due to their high temporal resolution, wide coverage, gratuity, and long observation period. However, owing to their low spatial resolution, the water boundary results are often blurred. It is difficult to extract water boundaries accurately. The subpixel mapping algorithm can solve this problem. In this article, Dongting Lake and its surroundings are adopted as the experimental area. The digital elevation model (DEM) is used to modify the subpixel/pixel spatial attraction model (SPSAM) mapping results. The proposed algorithm is referred to as the DEM-modified SPSAM (D-MSPSAM). Based on the visual results of the two sets of experiments, the modified results suitably maintain the spatial details of the water, and many of the underestimations caused by the similarity of the spectral characteristics of the surroundings to those of the water have been corrected. In this paper, the accuracy of Landsat-8 water extraction is used as a reference. Based on the quantitative results, the D-MSPSAM method has a higher extraction accuracy than the traditional threshold method, and the accuracies of the extraction for high water and low water have been increased by 3.56 percentage points and 2.77 percentage points, respectively. Furthermore, these results also confirm the potential application of DEM data for flood submergence extraction and provide new ideas for the improvement of the subpixel mapping model. The proposed method can accurately generate water distribution maps in a practical and economical way.
... It is used for various types of tasks and applications such as filling missing data (also known as image inpainting) and generating high-resolution images (e.g., pan-sharpening [11] and super-resolution [12]). It can include filtering approaches such as fusing information within a local window using methods such as interpolation [13,14], maximum a posteriori (MAP), Bayesian model, Markov random fields (MRFs), and Neural Networks (NN) [4,12,[15][16][17][18]. Although spatial-spectral fusion is efficient, it is not able to incorporate information from temporal images, which produce dramatic radiometric differences such as those introduced by meteorological, phenological, or ecological changes. ...
Remote sensing images and techniques are powerful tools to investigate earth’s surface. Data quality is the key to enhance remote sensing applications and obtaining clear and noise-free set of data is very difficult in most situations due to the varying acquisition (e.g., atmosphere and season), sensor and platform (e.g., satellite angles and sensor characteristics) conditions. With the increasing development of satellites, nowadays Terabytes of remote sensing images can be acquired every day. Therefore, information and data fusion can be particularly important in the remote sensing community. The fusion integrates data from various sources acquired asynchronously for information extraction, analysis, and quality improvement. In this chapter, we aim to discuss the theory of spatiotemporal fusion by investigating previous works, in addition to describing the basic concepts and some of its applications by summarizing our prior and ongoing works.
... To address this problem, researchers have employed the SRM method to extract the water body. SRM is intended to generate a land-cover map with higher spatial resolution than the input data [37,38]. From this perspective, SRM can be utilized to generate water body maps of finer spatial resolution from the Sentinel-3 images. ...
Water body mapping is significant for water resource management. In the view of 21 spectral bands and a short revisit time of no more than two days, a Sentinel-3 OLCI (Ocean and Land Colour Instrument) image could be the optimum data source in the near-real-time mapping of water bodies. However, the image is often limited by its low spatial resolution in practice. Super-resolution mapping (SRM) is a good solution to generate finer spatial resolution maps than the input data allows. In this paper, a multiscale spatiotemporal super-resolution mapping (MSST_SRM) method for water bodies is proposed, particularly for Sentinel-3 OLCI images. The proposed MSST_SRM method employs the integrated Normalized Difference Water Index (NDWI) images calculated from four near-infrared (NIR) bands and Green Band 6 of the Sentinel-3 OLCI image as input data and combined the spectral, multispatial, and temporal terms into one objective function to generate a fine water body map. Two experiments in the Tibet Plate and Daye lakes were employed to test the effectiveness of the MSST_SRM method. Results revealed that by using multiscale spatial dependence under the framework of spatiotemporal super-resolution Mapping, MSST_SRM could generate finer water body maps than the hard classification method and the other three SRM-based methods. Therefore, the proposed MSST_SRM method shows marked efficiency and potential in water body mapping using Sentinel-3 OLCI images.
... Land use and land cover are key factors in understanding important issues such as climate change, natural resource management, and urban and regional planning [13]. A land cover map describes the physical and biological cover of an area (e.g., grassland or water), whereas a land use map reveals human activity (e.g., infrastructure) [14][15][16]. Defining classes is required to generate a land cover and land use map. The detailed definitions of land use categories and building categories are provided in Table 1. ...
The utilization of urban land use maps can reveal the patterns of human behavior through the extraction of the socioeconomic and demographic characteristics of urban land use. Remote sensing that holds detailed and abundant information on spectral, textual, contextual, and spatial configurations is crucial to obtaining land use maps that reveal changes in the urban environment. However, social sensing is essential to revealing the socioeconomic and demographic characteristics of urban land use. This data mining approach is related to data cleaning/outlier removal and machine learning, and is used to achieve land use classification from remote and social sensing data. In bicycle and taxi density maps, the daytime destination and nighttime origin density reflects work-related land uses, including commercial and industrial areas. By contrast, the nighttime destination and daytime origin density pattern captures the pattern of residential areas. The accuracy assessment of land use classified maps shows that the integration of remote and social sensing, using the decision tree and random forest methods, yields accuracies of 83% and 86%, respectively. Thus, this approach facilitates an accurate urban land use classification. Urban land use identification can aid policy makers in linking human activities to the socioeconomic consequences of different urban land uses.
... MRF has conventionally been widely used to integrate context during image classification. For example, classification of hyperspectral images (Cao et al. 2018), image denoising (Cao et al. 2011), mapping of distribution of classes in sub-pixel classification (Kasetkasem et al. 2005), super-resolution mapping using MRF integrated with RF (Sanpayao et al. 2017) and spatial-temporal image classification (Jeon and Landgrebe 1992;Solberg et al. 1996;Melgani and Serpico 2003;Liu et al. 2006Liu et al. , 2008Moser and Serpico 2011). In Forest applications, MRFs has been used to map forest Li et al. (2014), monitor forest encroachment (Tiwari et al. 2016), and forest map revision (Solberg 1999). ...
The Kenyan coast is constantly under persistent cloud cover which hinders mapping using optical images.
Up-to-date land-cover information in such areas is sometimes missing from national mapping initiatives.
This study uses a computed composite image based on a mean of cloud and shadow free Function of Mask masked multi-temporal Landsat 8 images acquired during long-dry season in a pilot area. We test the effectiveness of the composite to map mangrove forest using random forest (RF) and support vector machines (SVM) machine learning algorithms integrated with context from Markov random fields (MRF(s)).
MRFs was chosen because it is computationally efficient hence can be scaled out nationally. The MRF frameworks are compared to pixel-based classification using threefold independent validation samples. SVM–MRFs and RF–MRFs methods have the highest overall accuracy compared to pixel-based classification.
However, visual assessment of predicted land-cover using aerial photograghs established that SVM–MRFs framework corresponded well to land-cover in the study area. This framework also managed to map classes with limited ground reference data better than RF–MRFs. Generally, context in both techniques played a discriminative role especially in heterogeneous regions. Therefore, scaling out this approaches would go a long way in generating mangrove forest map inventory in persistent cloud cover regions which is useful for land-based emission estimation.
... The SPM analysis typically uses the sub-pixel scale land cover information contained in class fraction images to locate the classes geographically in the area represented by CR pixels. A range of SPM algorithms has been developed [7][8][9][10][11][12] and the approach has been applied to data sets ranging from a single, mono-temporal, CR image to spatio-temporal SPM that uses a time series of CR images [13,14]. In the latter, the analysis may also be enhanced through the integration of information from a few fine spatial resolution (FR) land cover maps, if available, which enable the generation of FR land cover maps at the temporal frequency of the CR imagery [15]. ...
The generation of land cover maps with both fine spatial and temporal resolution would aid the monitoring of change on the Earth’s surface. Spatio-temporal sub-pixel land cover mapping (STSPM) uses a few fine spatial resolution (FR) maps and a time series of coarse spatial resolution (CR) remote sensing images as input to generate FR land cover maps with a temporal frequency of the CR data set. Traditional STSPM selects spatially adjacent FR pixels within a local window as neighborhoods to model the land cover spatial dependence, which can be a source of error and uncertainty in the maps generated by the analysis. This paper proposes a new STSPM using FR remote sensing images that pre- and/or post-date the CR image as ancillary data to enhance the quality of the FR map outputs. Spectrally similar pixels within the locality of a target FR pixel in the ancillary data are likely to represent the same land cover class and hence such same-class pixels can provide spatial information to aid the analysis. Experimental results showed that the proposed STSPM predicted land cover maps more accurately than two comparative state-of-the-art STSPM algorithms.
... Readers are referred to (Mertens 2008) for more details of SRM. So far a number of SRM models have been developed, such as subpixel/pixel spatial attraction models (Mertens et al. 2006), Hopfield neural networks (Tatem et al. 2002) and Markov random field-based model (Kasetkasem, Arora, and Varshney 2005). In this work the Artificial Neural Networks Predicted Wavelet Transform (ANN WT) method (Mertens et al. 2004) is employed. ...
This paper provides a study of the changes in land use in urban environments in two cities, Wuhan, China and western Sydney in Australia. Since mixed pixels are a characteristic of medium resolution images such as Landsat, when used for the classification of urban areas, due to changes in urban ground cover within a pixel, Multiple Endmember Spectral Mixture Analysis (MESMA) together with Super-Resolution Mapping (SRM) are employed to derive class fractions to generate classification maps at a higher spatial resolution using an Artificial Neural Network (ANN) predicted Wavelet method. Landsat images over the two cities for a 30-year period, are classified in terms of vegetation, buildings, soil and water. The classifications are then processed using Indifrag software to assess the levels of fragmentation caused by changes in the areas of buildings, vegetation, water and soil over the 30 years. The extents of fragmentation of vegetation, buildings, water and soil for the two cities are compared, while the percentages of vegetation are compared with recommended percentages of green space for urban areas for the benefit of health and well-being of inhabitants. Changes in Ecosystem Service Values (ESVs) resulting from the urbanization have been assessed for Wuhan and Sydney. The UN Sustainable Development Goals (SDG) for urban areas are being assessed by researchers to better understand how to achieve the sustainability of cities.
Subpixel mapping (SPM) aims to reconstruct a subpixel-level class distribution map from the pixel-level abundance maps, which is an under-determined problem that has non-unique solutions. To address this, the spatiotemporal SPM uses the abundance, spatial, and temporal constraints to reduce the uncertainty of the mapping solutions, so the spatiotemporal SPM is essentially a constrained optimization problem. However, it is hard to find the optimal weighting parameters to combine the three joint constraints. In addition, the existing spatiotemporal SPM methods mainly use the temporal information either for the unchanged subpixels detection or for the subpixel classification, which is insufficient in the utilization of the temporal information. In this paper, a novel multiobjective optimization-based spatiotemporal SPM algorithm (STSPM_MO) is proposed. STSPM_MO is composed of an unchanged subpixels detection stage and a multiobjective spatiotemporal mapping stage. In the former stage, the historical thematic map is used for identifying the unchanged subpixels. In the latter stage, the historical thematic map is further used for providing the temporal dependence, so that the temporal information can be more fully utilized. Moreover, to solve the constrained optimization problem of the spatiotemporal SPM, the abundance, spatial and temporal constraints are modeled as three objectives and are dynamically fused through the subfitness-based multiobjective evolution, to generate the optimal subpixel classification map. Both synthetic and real-data experiments have been conducted and the results show the proposed method and its two variants are superior, stable, and effective.
The ubiquity of mixed pixels in hyperspectral images makes it difficult for traditional classification techniques to determine the spatial distribution of land cover classes accurately. Subpixel mapping (SPM) technology is an effective method to solve this problem. Aiming at taking the multiple scales and the spatial features into account, an SPM method based on multi-scale and multi-feature (MSMF) is proposed, so as to effectively improve the accuracy of SPM. Firstly, the maximum linearization index method of the non-redundant complete straight-line set is designed to identify the linear distribution feature of land-cover classes. And then, different methods are applied to different spatial features and unified together finally, where the template matching iterative exchange is used for the linear distribution classes, and the multiscale spatial dependence iterative exchange method combined with area perimeter is used for the planar distribution classes. Experiments on three remote sensing images are carried out to evaluate the performance of MSMF. The results show that the proposed method can effectively improve the accuracy of SPM.
In order to locate forest fire smoke more precisely and expand existing forest fire monitoring methods, this research employed Himawari-8 data with a sub-pixel positioning concept in smoke detection. In this study, Himawari-8 data of forest fire smoke in Xichang and Linzhi were selected. An improved sub-pixel mapping method based on random forest results was proposed to realize the identification and sub-pixel positioning of smoke. More spatial details of forest fire smoke were restored in the final results. The continuous monitoring of smoke indicated the dynamic changes therein. The accuracy evaluation of smoke detection was realized using a confusion matrix. Based on the improved sub-pixel mapping method, the overall accuracies were 87.95% and 86.32%. Compared with the raw images, the smoke contours of the improved sub-pixel mapping results were clearer and smoother. The improved sub-pixel mapping method outperforms traditional classification methods in locating smoke range. Moreover, it especially made a breakthrough in the limitations of the pixel scale and in realizing sub-pixel positioning. Compared with the results of the classic PSA method, there were fewer “spots” and “holes” after correction. The final results of this study show higher accuracies of smoke discrimination, with it becoming the basis for another method of forest fire monitoring.
Remote sensing (RS) images are considered to be reflections of the real world. However, RS images often suffered from low resolution, making further research difficult to follow. Although super resolution (SR) techniques based on deep learning have achieved considerable breakthroughs, they show limited performance when dealing with low-quality RS images with complicated backgrounds; for instance, the SR results tend to loss details and have undesired structural distortion. Thus, this paper proposes an innovative dual-branch attention network (DBAN) to produce sufficient details and preserve clear structural information for SR results of RS images. It consists of two components: a feature extraction branch and a high-frequency information learning branch. The features extraction branch, formed as a densely residual structure, combines a series of dual attention blocks that are designed to exploit valid features from different dimensions, and then all these multi-scale features are reused through a global concatenation. The high-frequency information extraction branch, incorporating noise removing units (NRU) and high-frequency attention units (HFU), is responsible for producing the high-frequency features without noise, which enables DBAN to handle the problem of structural distortion. Meanwhile, a composite loss function based on a Laplacian pyramid is proposed to maximize the structural similarity between reconstruction results and real high-resolution RS images. The proposed network is efficient and lightweight because of its strong and effective attention to feature learning. Experimental results on three open-source RS image datasets and the JiLin-01 dataset demonstrate the effectiveness of our DBAN where higher accuracy over state-of-the-art methods in super-resolving complicated images was achieved.
Bayesian image processing has progressively increased in importance in various fields of the natural sciences. It utilizes prior knowledge and forward models of the observational processes through Bayes' theorem, enabling the accurate estimation of model parameters that represent the physical quantities of the target. Moreover, using hyperparameter estimation, we can determine the hidden physical parameters that govern the processes in and the structure of the target and sensing systems, such as the spatial continuity of the model parameters and the magnitude of the observational noise. Such a general framework, which uses Bayesian estimation to understand the essential physics of a target system, can be called Bayesian sensing. This paper discusses the physical meaning of and the mechanism underlying Bayesian sensing using the concept of resolution in the spatial-inversion problem. The spatial resolution of the model parameters can be mapped using a resolution matrix, more rigorously, a model resolution matrix defined as a linear mapping from the true model parameters to the recovered model parameters. We formulate the resolution matrix for Bayesian image processing and also show that in terms of resolution, the optimal hyperparameters are obtained from internally consistent equations that connect the estimated optimal hyperparameters with the actual ones calculated from the estimated model parameters. In addition, we show the equivalence of the internally consistent equations to the expectation-maximization (EM) algorithm, and we formulate the confidence intervals for the estimated hyperparameters, which indicate the general convergence of the hyperparameter estimates. We also show the effectiveness of the proposed method by performing synthetic numerical tests for two inversion-problem settings: linear travel-time seismic tomography and image deblurring. The resulting equations can contribute to understanding the hidden physical processes in and the structure of the target and observation systems for various problems.
TAIMA 2022, onzième édition de la conférence internationale sur le Traitement et l’Analyse de l’Information : Méthodes et Applications, se tient du 28 Mai au 02 Juin 2022 à Hammamet, Tunisie. Elle a pour objectif de faire le point sur l’état de l’art et les dernières avancées dans les technologies, méthodologies et applications potentielles qui se dégagent dans le domaine du traitement et de l’analyse d’information. La conférence TAIMA rassemble la communauté scientifique du traitement, de l’analyse, de l’apprentissage pour les signaux et images autour d’experts internationaux de renom sur ces domaines.
Global land cover mapping activities are of great important for retrieving the environment we are living in. However, the trade-off between spatial and temporal resolution makes it difficult to obtain the continuous fine-scale land cover product for detail and frequent land surface analysis. To overcome the difficulty, this study proposed a modified maximum a posteriori (MAP) based spatiotemporal sub-pixel mapping (SSM), for a long-term and fine-scale land cover mapping. The proposed approach consists of three parts, i.e., data fidelity term, spatial regularization term, and enhanced temporal regularization term, which can make use of historical fine-scale image for a better super-resolution of the current coarse image. One regional experiment was implemented for the algorithm validation, and one real experiment was investigated, by generating continuous land cover mapping products (2010–2015) at Landsat-like spatial resolution from time series of MODIS images. Compared with traditional methods, the proposed approach reconstructed more pleasant spatial details and had greater quantitative accuracy, outperforming the state-of-the-art spatiotemporal sub-pixel mapping method by averagely 2.58% for all the inspected dates in OA metric. This study not only provides a generalized research framework for generating continuous fine-scale land cover product from daily available MODIS images, but also validates the usability of the time series products for detail land cover monitoring, and reveals a finding that the study area of Wuhan is a well practitioner of the sustainable development policy during its urbanization process.
Conventional subpixel mapping (SPM) is performed based on the abundance maps obtained by spectral unmixing (SU), to interpret the mixed pixels and improve the mapping resolution for hyperspectral remote sensing imagery. However, the SU and SPM tasks are separately conducted, so that the unmixing error is propagated to the SPM, and the mapping result is strongly reliant on the quality of the abundance maps. In this paper, a novel joint SPM and SU framework (MO_SUSM) based on multiobjective optimization is proposed to simultaneously perform unmixing and mapping. Specifically, the multiobjective joint optimization model with data fidelity term and Laplacian prior term is constructed for the SU and SPM. For the data fidelity term, since the unmixing result can be recovered by downsampling the mapping result, the unmixing model is joined with the mapping model by the downsampling matrix, so that the reconstruction errors of the unmixing and mapping results can be minimized together. Meanwhile, the Laplacian prior term is used to maximize the spatial dependence of the mapping result and provide the spatial constraint for the SU. In addition, the multiobjective optimization algorithm with local search is designed to search for the optimal unmixing and mapping results that can balance the objective terms. Since the two objective terms are dynamically integrated during the optimization, there is no need to set sensitive weights for the objectives combination. Four experiments were conducted on hyperspectral images of various data sources, including ground, airborne, and satellite images. The unmixing results show that MO_SUSM can reduce the unmixing error, and can improve the quality of the abundance maps. The mapping results show that MO_SUSM can alleviate the dependence of SPM on the abundance maps, and can improve the mapping accuracy.
The mixed pixel problem is common in remote sensing. A soft classification can generate land cover class fraction images that illustrate the areal proportions of the various land cover classes within pixels. The spatial distribution of land cover classes within each mixed pixel is, however, not represented. Super-resolution land cover mapping (SRM) is a technique to predict the spatial distribution of land cover classes within the mixed pixel using fraction images as input. Spatial-temporal SRM (STSRM) extends the basic SRM to include a temporal dimension by using a finer-spatial resolution land cover map that pre- or postdates the image acquisition time as ancillary data. Traditional STSRM methods often use one land cover map as the constraint, but neglect the majority of available land cover maps acquired at different dates and of the same scene in reconstructing a full state trajectory of land cover changes when applying STSRM to time-series data. In addition, the STSRM methods define the temporal dependence globally, and neglect the spatial variation of land cover temporal dependence intensity within images. A novel local STSRM (LSTSRM) is proposed in this paper. LSTSRM incorporates more than one available land cover map to constrain the solution, and develops a local temporal dependence model, in which the temporal dependence intensity may vary spatially. The results show that LSTSRM can eliminate speckle-like artifacts and reconstruct the spatial patterns of land cover patches in the resulting maps, and increase the overall accuracy compared with other STSRM methods.
Urban forests play a significant role in improving air quality and climate protection, energy saving, recreation and human connection with nature. To maximize on the mentioned benefits, urban trees inventory is necessary to identify tree location, gain species information and their spatial distributions. Urban areas do have a mixed environment of land cover. In addition, traditional techniques like ground survey, aerial photography are time-consuming, costly and limited. This study researched the classification of urban tree species using maximum likelihood classification and support vector machine on the WorldView-2 satellite image. This entailed using object and pixels in both methods. The main objective of this study was to classify urban tree species using Maximum Likelihood Classification (MLC) and Support Vector Machine (SVM). Object based and pixel based analysis was used in both methods. MLC performed better than SVM in both object and pixel based calssification. MLC pixel based overall accuracy was 66.93% with Kappa of 0.54 and 51.24% with Kappa of 0.36 for SVM pixel based. MLC object based overall accuracy was 71.17% with Kappa of 0.66 and 44.62% with Kappa of 0.31 for SVM object based analysis. Even though MLC perfoms better than SVM, the accuracy is still low compared to generally accepted accuracy. This indicates that both methods are still not satisfactory techniques of classifying high resolution images for Delft city. MLC, however, has been used for many years in image classification. It is straightforward and does not require extreme expert skills to apply. MLC algorithm can be found in most of the remote sensing application software. Examples of this software are ERDAS, ENVI and ILWIS among others. This makes MLC an easy available method for classification. MLC pixel based classification is effective in classifying medium and large trees (e.g. Plantanus Spp. and Fagus Spp.). MLC relies on mean and covariance of samples hence calculation of covariance matrix in small tree crowns (> 10 pixels) could not be determined. Class separability using J-M distance measure and NDVI mean and variation evaluation were the same. This gives possibilty of use of NDVI in separating tree species. SVM does not operate based on data distribution making it applicable to any type of data (i.e. normal or non-normal distribution. Its performance relies on kernel parameters. In this study, C value of 5 and value 5 for δ were used. Experimenting on optimum parameters values of C and δ can give satisfactory classification results. Though the study area is in The Netherlands, classification of tree species using MLC and SVM brings up possibilities to apply the same approach in other urban areas and to other tree species.
Subpixel mapping (SPM) is a useful technique that can interpret the spatial distribution inside mixed pixels and produce a finer-resolution classification map for hyperspectral remote-sensing imagery. However, SPM is essentially an ill-posed problem that requires additional information to produce the unique solution. The limited information of a single image is insufficient to make the mapping problem well posed, whereas the complementary spatial information of multiple shifted images is able to reduce the uncertainty and generate an accurate map. The maximum a posteriori model is a feasible way to incorporate auxiliary information for SPM with multiple shifted images, but it introduces a sensitive regularization parameter, which is difficult to preset. Furthermore, the fixed parameter in the iterations influences the incorporation of the multiple images and the spatial prior. In this article, to address these issues, a multiobjective SPM framework for use with multiple shifted hyperspectral images (MOMSM) is proposed. In the proposed algorithm, a multiobjective model consisting of two objective functions, i.e., data fidelity and spatial prior terms, is constructed to transform the SPM into a multiobjective optimization problem, to get rid of the sensitive regularization parameter. To simultaneously optimize the two objective functions, a multiobjective memetic algorithm with a local search operator and an adaptive global replacement strategy is proposed. The multiple images and spatial information can be dynamically fused and the optimal mapping solution with a good balance between the two objectives can be finally obtained. Experiments conducted on both synthetic and real data sets confirm that the proposed method outperforms the other tested SPM algorithms.
Remote sensing images contain abundant land cover information. Due to the complex nature of land cover, however, mixed pixels exist widely in remote sensing images. Sub-pixel mapping (SPM) is a technique for predicting the spatial distribution of land cover classes within mixed pixels. As an ill-posed inverse problem, the uncertainty of prediction cannot be eliminated and hinders the production of accurate sub-pixel maps. In contrast to conventional methods that use continuous geospatial information (e.g., images) to enhance SPM, in this paper, a SPM method with point constraints into SPM is proposed. The method of fusing point constraints is implemented based on the pixel swapping algorithm (PSA) and utilizes the auxiliary point information to reduce the uncertainty in the SPM process and increase map accuracy. The point data are incorporated into both the initialization and optimization processes of PSA. Experiments were performed on three images to validate the proposed method. The influences of the performances were also investigated under different numbers of point data, different spatial characters of land cover and different zoom factors. The results show that by using the point data, the proposed SPM method can separate more small-sized targets from aggregated artifacts and the accuracies are increased obviously. The proposed method is also more accurate than the advanced radial basis function interpolation-based method. The advantage of using point data is more evident when the point data size and scale factor are large and the spatial autocorrelation of the land cover is small. As the amount of point data increases, however, the increase in accuracy becomes less noticeable. Furthermore, the SPM accuracy can still be increased even if the point data and coarse proportions contain errors.
Heterogeneous crop identification has been the subject of much concern, since smallholder farms less than 1 ha are the main agricultural form in many areas, especially China. Remote sensing with high spectral and spatial resolutions via aerial platforms such as unmanned aerial vehicles (UAV) provides a potential alternative technique for the monitoring of heterogeneous crops in smallholder agriculture. Although this new type of remote sensing data with high spectral and spatial resolutions provides the possibility of fine classification, it also brings some challenges, such as bands contaminated with severe noise, the nonuniform distribution of the discriminative spectral information, and the spectral variability of crops. In this study, we attempted to resolve these problems by developing a robust spectral-spatial agricultural crop mapping method based on conditional random fields (SCRF), which learns the sensitive spectral information of the crops by a spectrally weighted kernel, and uses the spatial interaction of pixels to improve the classification performance. Data from a manned aircraft platform and a UAV platform were chosen to validate the effectiveness of the proposed algorithm. The experimental results showed that the proposed algorithm can effectively use the relative utility of each spectral band to detect the bands contaminated with severe noise, and it uses the spectrally weighted kernel to consider the sensitive spectral information of the crops. The algorithm with only a spectrally weighted kernel showed an improvement of more than 4% over the classical support vector machine and random forest methods. Moreover, the spatial information was proved to be of crucial importance for crop classification, and both the object-oriented method and the proposed SCRF method can improve the classification performance in terms of both visualization and the quantitative metrics by considering the spatial information. Compared with the object-oriented method, SCRF can deliver a better classification performance, with an accuracy improvement of more than 2%.
Although accurate training and initialization information is difficult to acquire, unsupervised hyperspectral subpixel mapping (SPM) without relying on this predefined information is an insufficiently addressed research issue. This letter presents a novel Bayesian approach for unsupervised SPM of hyperspectral imagery (HSI) based on the Markov random field (MRF) and a band-weighted discrete spectral mixture model (BDSMM), with the following key characteristics. First, this is an unsupervised approach that allows adjustment of abundance and endmember information adaptively for less relying on algorithm initialization. Second, this approach consists of the BDSMM for accommodating the noise heterogeneity and the hidden label field of subpixels in HSI. The BDSMM also integrates SPM into the spectral mixture analysis and allows enhanced SPM by fully exploring the endmember-abundance patterns in HSI. Third, the MRF and BDSMM are integrated into a Bayesian framework to use both the spatial and spectral information efficiently, and an expectation-maximization (EM) approach is designed to solve the model by iteratively estimating the endmembers and the label field. Experiments on both simulated and real HSI demonstrate that the proposed algorithm can yield better performance than traditional methods.
Due to the instantaneous field-of-view (IFOV) of the sensor and diversity of land cover types, some pixels, usually named mixed pixels, contain more than one land cover type. Soft classification can predict the portion of each land cover type in mixed pixels in the absence of spatial distribution. The spatial distribution information in mixed pixels can be solved by super resolution mapping (SRM). Typically, SRM involves two steps: soft class value estimation, which is similar to the image super resolution of image restoration, and land cover allocation. A new SRM approach utilizes a deep image prior (DIP) strategy combined with a super resolution convolutional neural network (SRCNN) to estimate fine resolution fraction images for each land cover type; then, a simple and efficient classifier is used to allocate subpixel land cover types under the constraint of the generated fine fraction images. The proposed approach can use prior information of input images to update network parameters and no longer require training data. Experiments on three different cases demonstrate that the subpixel classification accuracy of the proposed DIP-based SRM approach is significantly better than the three conventional SRM approaches and a transfer learning-based neural network SRM approach. In
addition, the DIP-SRM approach performs very robustly about small-area objects within multiple land cover types and significantly reduces soft classification uncertainty. The results of this paper provide an extension for utilizing SRCNN to address SRM issues in hyperspectral images.
Many techniques intended to estimate land coverage of multiple categories occupied within each pixel from such coarse resolution data have been proposed. However, in traditional unmixing studies with coarse resolution imagery such as Advanced Very High Resolution Radiometer (AVHRR) data, it is assumed that only a few endmembers exist throughout an entire image. Therefore, it is essential to evaluate how well an unmixing method would work for various categories within pixels of coarse resolution images. In this study, the land coverage of eight classes in National Oceanic and Atmospheric Administration (NOAA) AVHRR imagery by using finer resolution Landsat Thematic Mapper (TM) imagery was estimated, and the accuracy of these estimated classes was evaluated. The results suggest that this method may be generally useful for comparing multi-spectral images in space and time.
In this paper we consider how we may determine the relative proportions of ground cover components in a mixed pixel. We assume the usual linear model for signal mixing and examine a number of methods, closely related, for estimating the proportions. We also show how the precision of our estimates can be defined. We introduce a new estimator which is based on regularisation principles and which produces a smoother set of images than other methods, and gives more accurate estimates. The methods are compared on a simulated data set.
This paper transmits a FORTRAN-IV coding of the fuzzy c-means (FCM) clustering program. The FCM program is applicable to a wide variety of geostatistical data analysis problems. This program generates fuzzy partitions and prototypes for any set of numerical data. These partitions are useful for corroborating known substructures or suggesting substructure in unexplored data. The clustering criterion used to aggregate subsets is a generalized least-squares objective function. Features of this program include a choice of three norms (Euclidean, Diagonal, or Mahalonobis), an adjustable weighting factor that essentially controls sensitivity to noise, acceptance of variable numbers of clusters, and outputs that include several measures of cluster validity.
All of us frequently encounter decision-making problems in every day life. Based on our observations regarding a certain phenomenon, we need to select a particular course of action from a set of possible options. This problem involving a single decision maker is typically a difficult one. Decision making in large-scale systems consisting of multiple decision makers is an even more challenging problem. Group decision-making structures are found in many real world situations. Application areas include financial institutions, air-traffic control, oil exploration, medical diagnosis, military command and control, electric power networks, weather prediction, and industrial organizations. For example, a medical doctor may order multiple diagnostic tests and seek additional peer opinions before a major surgical procedure is carried out, or a military commander may use data from radar and IR sensors along with intelligence information while deciding whether or not to launch an offensive. In many applications, multiple decision makers arise naturally, e.g., managers in an industrial organization. In many other applications, additional decision makers are employed to improve system performance. For example, deployment of multiple sensors for signal detection in a military surveillance application improves system survivability, results in improved detection performance or in a shorter decision time to attain a prespecified performance level, and may provide increased coverage in terms of surveillance region and number of targets.
Land cover class composition of remotely sensed image pixels can be
estimated using soft classification techniques increasingly available in many GIS
packages. However, their output provides no indication of how such classes are
distributed spatially within the instantaneous field of view represented by the
pixel. Techniques that attempt to provide an improved spatial representation of
land cover have been developed, but not tested on the difficult task of mapping
from real satellite imagery. The authors investigated the use of a Hopfield neural
network technique to map the spatial distributions of classes reliably using
information of pixel composition determined from soft classification previously.
The approach involved designing the energy function to produce a ‘best guess’
prediction of the spatial distribution of class components in each pixel. In previous
studies, the authors described the application of the technique to target identification,
pattern prediction and land cover mapping at the sub-pixel scale, but only
for simulated imagery.We now show how the approach can be applied to Landsat
Thematic Mapper (TM) agriculture imagery to derive accurate estimates of land
cover and reduce the uncertainty inherent in such imagery. The technique was
applied to Landsat TM imagery of small-scale agriculture in Greece and largescale
agriculture near Leicester, UK. The resultant maps provided an accurate
and improved representation of the land covers studied, with RMS errors for the
Landsat imagery of the order of 0.1 in the new fine resolution map recorded. The
results showed that the neural network represents a simple efficient tool formapping land cover from operational satellite sensor imagery and can deliver
requisite results and improvements over traditional techniques for the GIS analysis
of practical remotely sensed imagery at the sub pixel scale.
In remotely sensed images, mixed pixels will always be present. Soft classification defines the membership degree of these pixels for the different land cover classes. Sub-pixel mapping is a technique designed to use the information contained in these mixed pixels to obtain a sharpened image. Pixels are divided into sub-pixels, representing the land cover class fractions. Genetic algorithms combined with the assumption of spatial dependence assign a location to every sub-pixel. The algorithm was tested on synthetic and degraded real imagery. Obtained accuracy measures were higher compared with conventional hard classifications.
Fuzzy classification techniques have been developed recently to estimate the class composition of image pixels, but their output provides no indication of how these classes are distributed spatially within the instantaneous field of view represented by the pixel. As such, while the accuracy of land cover target identification has been improved using fuzzy classification, it remains for robust techniques that provide better spatial representation of land cover to be developed. Such techniques could provide more accurate land cover metrics for determining social or environmental policy, for example. The use of a Hopfield neural network to map the spatial distribution of classes more reliably using prior information of pixel composition determined from fuzzy classification was investigated. An approach was adopted that used the output from a fuzzy classification to constrain a Hopfield neural network formulated as an energy minimization tool. The network converges to a minimum of an energy function, defined as a goal and several constraints. Extracting the spatial distribution of target class components within each pixel was, therefore, formulated as a constraint satisfaction problem with an optimal solution determined by the minimum of the energy function. This energy minimum represents a "best guess" map of the spatial distribution of class components in each pixel. The technique was applied to both synthetic and simulated Landsat TM imagery, and the resultant maps provided an accurate and improved representation of the land covers studied, with root mean square errors (RMSEs) for Landsat imagery of the order of 0.09 pixels in the new fine resolution image recorded.
An unsupervised segmentation approach to classification of
multispectral image is suggested here in Markov random field (MRF) frame
work. This work generalizes the work of Sarkar et al. (2000) on gray
value images for multispectral images and is extended for landuse
classification. The essence of this approach is based on capturing
intrinsic characters of tonal and textural regions of any multispectral
image. The approach takes an initially oversegmented image and the
original. multispectral image as the input and defines a MRF over region
adjacency graph (RAG) of the initially segmented regions. Energy
function minimization associated with the MRF is carried out by applying
a multivariate statistical test. A cluster validation scheme is outlined
after obtaining optimal segmentation. Quantitative evaluation of
classification accuracy of test data for three illustrations are shown
and compared with conventional maximum likelihood procedure. Comparison
of the proposed methodology with a recent work of texture segmentation
in the literature has also been provided. The findings of the proposed
method are found to be encouraging
Mixture modeling is becoming an increasingly important tool in the
remote sensing community as researchers attempt to resolve subpixel,
area information. This paper compares a well-established technique,
linear spectral mixture models (LSMM), with a much newer idea based on
data selection, support vector machines (SVM). It is shown that the
constrained least squares LSMM is equivalent to the linear SVM, which
relies on proving that the LSMM algorithm possesses the “maximum
margin” property. This in turn shows that the LSMM algorithm can
be derived from the same optimality conditions as the linear SVM, which
provides important insights about the role of the bias term and rank
deficiency in the pure pixel matrix within the LSMM algorithm. It also
highlights one of the main advantages for using the linear SVM algorithm
in that it performs automatic “pure pixel” selection from a
much larger database. In addition, extensions to the basic SVM algorithm
allow the technique to be applied to data sets that exhibit spectral
confusion (overlapping sets of pure pixels) and to data sets that have
nonlinear mixture regions. Several illustrative examples, based on an
area-labeled Landsat dataset, are used to demonstrate the potential of
this approach
A general model for multisource classification of remotely sensed
data based on Markov random fields (MRF) is proposed. A specific model
for fusion of optical images, synthetic aperture radar (SAR) images, and
GIS (geographic information systems) ground cover data is presented in
detail and tested. The MRF model exploits spatial class dependencies
(spatial context) between neighboring pixels in an image, and temporal
class dependencies between different images of the same scene. By
including the temporal aspect of the data, the proposed model is
suitable for detection of class changes between the acquisition dates of
different images. The performance of the proposed model is investigated
by fusing Landsat TM images, multitemporal ERS-1 SAR images, and GIS
ground-cover maps for land-use classification, and on agricultural crop
classification based on Landsat TM images, multipolarization SAR images,
and GIS crop field border maps. The performance of the MRF model is
compared to a simpler reference fusion model. On an average, the MRF
model results in slightly higher (2%) classification accuracy when the
same data is used as input to the two models. When GIS field border data
is included in the MRF model, the classification accuracy of the MRF
model improves by 8%. For change detection in agricultural areas, 75% of
the actual class changes are detected by the MRF model, compared to 62%
for the reference model. Based on the well-founded theoretical basis of
Markov random field models for classification tasks and the encouraging
experimental results in our small-scale study, the authors conclude that
the proposed MRF model is useful for classification of multisource
satellite imagery
An adaptive segmentation algorithm is developed which
simultaneously estimates the parameters of the underlying Gibbs random
field (GRF)and segments the noisy image corrupted by additive
independent Gaussian noise. The algorithm, which aims at obtaining the
maximum a posteriori (MAP) segmentation is a simulated annealing
algorithm that is interrupted at regular intervals for estimating the
GRF parameters. Maximum-likelihood (ML) estimates of the parameters
based on the current segmentation are used to obtain the next
segmentation. It is proven that the parameter estimates and the
segmentations converge in distribution to the ML estimate of the
parameters and the MAP segmentation with those parameter estimates,
respectively. Due to computational difficulties, however, only an
approximate version of the algorithm is implemented. The approximate
algorithm is applied on several two- and four-region images with
different noise levels and with first-order and second-order
neighborhoods
Landscape pattern represents a key variable in management and understanding of the environment, as well as driving many environmental models. Remote sensing can be used to provide information on the spatial pattern of land cover features, but analysis and classification of such imagery suffers from the problem of class mixing within pixels. Soft classification techniques can estimate the class composition of image pixels. However, their output provides no indication of how such classes are distributed spatially within the instantaneous field-of-view (IFOV) represented by the pixel. Techniques to provide an improved spatial representation of land cover targets larger than the size of a pixel have been developed. However, the mapping of subpixel scale land cover features has yet to be investigated. We recently described the application of a Hopfield neural network technique to super-resolution mapping of land cover features larger than a pixel, using information of pixel composition determined from soft classification, and now show how our approach can be extended in a new way to predict the spatial pattern of subpixel scale features. The network converges to a minimum of an energy function defined as a goal and several constraints. Prior information on the typical spatial arrangement of the particular land cover types is incorporated into the energy function as a semivariance constraint. This produces a prediction of the spatial pattern of the land cover in question, at the subpixel scale. The technique is applied to synthetic and simulated Landsat Thematic Mapper (TM) imagery, and compared to results of an existing super-resolution target identification technique. Results show that the new approach represents a simple, robust, and efficient tool for super-resolution land cover pattern prediction from remotely sensed imagery.
A method was developed to transform a soft land cover classification into hard land cover classes at the sub-pixel scale for subsequent per-field classification. First, image pixels were segmented using vector boundaries. Second, the pixel segments (ranked by area) were labelled with a land cover class (ranked by class typicality). Third, a hard per-field classification was generated by examining each polygon (representing a land cover parcel, or field) in its entirety (by grouping the fragments of the polygon contained within different image pixels) and assigning to it the modal land cover class. The accuracy of this technique was considerably higher than that of both a corresponding hard per-pixel classification and a perfield classification based on hard per-pixel classified imagery.
Mixed pixels occur commonly in remotely-sensed imagery, especially those with a coarse spatial resolution. They are a problem in land-cover mapping applications since image classification routines assume ‘pure’ or homogeneous pixels. By unmixing a pixel into its component parts it is possible to enableinter alia more accurate estimation of the areal extent of different land cover classes. In this paper two approaches to estimating sub-pixel land cover composition are investigated. One is a linear mixture model the other is a regression model based on fuzzy membership functions. For both approaches significant correlation coefficients, all >0·7, between the actual and predicted proportion of a land cover type within a pixel were obtained. Additionally a case study is presented in which the accuracy of the estimation of tropical forest extent is increased significantly through the use of sub-pixel estimates of land-cover composition rather than a conventional image classification.
Fuzzy classification may be used to estimate the class composition of image pixels but it does not indicate how these classes are distributed spatially within the pixels. The potential to locate the distribution of classes more precisely through the use of a sharpening image was investigated. A sharpened fuzzy classification provided a visually improved representation of land cover and formed a more appropriate base for the derivation of quantitative measures of landscape mosaic than the fuzzy classification it was derived from.
Classification trees are a powerful alternative to more traditional approaches of land cover classification. Trees provide a hierarchical and nonlinear classification method and are suited to handling non-parametric training data as well as categorical or missing data. By revealing the predictive hierarchical structure of the independent variables, the tree allows for great flexibility in data analysis and interpretation. In this Letter, we compare a tree' s performance to that of a maximum likelihood classifier using a 1° by 1° global data sel. The tree's accuracy in classifying a validation dala set is comparable to that when using maximum likelihood (82 per cent). The tree also may be used to reduce the dimensionality of data sets and to find those metrics that are most useful for discriminating among cover types.
High-spatial resolution digital color-infrared aerial imagery of Syracuse, NY was analyzed to test methods for developing land cover classifications for an urban area. Five cover types were mapped: tree/shrub, grass/herbaceous, bare soil, water and impervious surface. Challenges in high-spatial resolution imagery such as shadow effect and similarity in spectral response between classes were found. Classification confusion among objects with similar spectral responses occurred between water and dark impervious surfaces, concrete and bare-soil, and grass/herbaceous and trees/shrub. Methods of incorporating texture, band ratios, masking of water objects, sieve functions, and majority filters were evaluated for their potential to improve the classification accuracy. After combining these various techniques, overall cover accuracy for the study area was 81.75%. Highest accuracies occurred for water (100%), tree/shrub (86.2%) and impervious surfaces (82.6%); lowest accuracy were for grass/herbaceous (69.3%) and bare soil (40.0%). Methods of improving cover map accuracy are discussed.
Mixed pixels result when the sensor's instantaneous field-of-view includes more than one land cover class on the ground. For mixed pixels, fuzzy classifiers can be used, which assign a pixel to several land cover classes in proportion to the area of the pixel that each class covers. These fraction values can be assigned to sub-pixels, based on the assumption of spatial dependence and the application of linear optimization techniques. A newly proposed sub-pixel mapping algorithm was first applied to a synthetic data set with a 1-km resolution, derived from a 20-m resolution image. This algorithm yielded land cover maps at 500, 200, and 100 m resolution with accuracies close to 89%. Subsequent mode filtering further increased these values. When applied to a real data set, the accuracy reached 78%. While this study suggests the potential of the proposed technique, there is still ample scope for improvements and extensions.
Remotely sensed data are an attractive source of land cover data over a wide range of spatial and temporal scales. The realisation of the full potential of remote sensing as a source of land cover data is, however, restricted by numerous factors. One commonly encountered problem is the presence of mixed pixels, which cannot be appropriately accommodated in conventional image classification techniques used in thematic mapping from remotely sensed data. This problem has generally been resolved through the adoption of a soft or fuzzy classification from which the fractional coverage of classes in the image pixels may be mapped. In this type of approach, the strength of membership, a pixel displays to a class, is used as a surrogate for the fractional coverage of that class. The accuracy of the resulting land cover representation is, therefore, dependent on the relationships between class membership strength and associated class fractional coverage. Since class membership can only be measured in relation to the classes defined in the training stage of the classification, untrained classes may influence the accuracy of the class composition estimation. For example, a pixel representing an area of an untrained class can only display membership to the trained classes. The effect of an untrained class on the accuracy of sub-pixel class composition estimation will depend on how the class membership strength is calculated. Here, the effect of untrained classes on sub-pixel land cover composition estimation using algorithms that produce relative and absolute measures of class membership was assessed. The algorithms investigated were the widely used fuzzy c-means (FCM) and its possibilistic counterpart, the possibilistic c-means (PCM), algorithms which derive relative and absolute measures of class membership strength, respectively. Both algorithms were able to provide accurate estimates of sub-pixel land cover composition. When all classes had been defined in training a classification, the FCM generally provided the most accurate class composition estimates. The presence of an untrained class, however, could substantially degrade the accuracy of the sub-pixel land cover composition estimates derived from the FCM but had no effect on those from the PCM. Since untrained classes are commonly encountered it may be more appropriate to use approaches such as the PCM in addition to, or instead of, the FCM to enhance the extraction of land cover information from remotely sensed data.
Despite the sizable achievements obtained, the use of soft classifiers is still limited by the lack of well-assessed and adequate methods for evaluating the accuracy of their outputs. This paper proposes a new method that uses the fuzzy set theory to extend the applicability of the traditional error matrix method to the evaluation of soft classifiers. It is designed to cope with those situations in which classification and/or reference data are expressed in multimembership form and the grades of membership represent different levels of approximation to intrinsically vague classes.
Sub-pixel mapping and sub-pixel sharpening are techniques for increasing the spatial resolution of sub-pixel image classifications. The proposed method makes use of wavelets and artificial neural networks. Wavelet multiresolution analysis facilitates the link between different resolution levels. In this work a higher resolution image is constructed after estimation of the detail wavelet coefficients with neural networks. Detail wavelet coefficients are used to synthesize the high-resolution approximation. The applied technique allows for both sub-pixel sharpening and sub-pixel mapping. An algorithm was developed on artificial imagery and tested on artificial as well as real synthetic imagery. The proposed method resulted in images with higher spatial resolution showing more spatial detail than the source imagery. Evaluation of the algorithm was performed both visually and quantitatively using established classification accuracy indices.
We make an analogy between images and statistical mechanics systems. Pixel gray levels and the presence and orientation of edges are viewed as states of atoms or molecules in a lattice-like physical system. The assignment of an energy function in the physical system determines its Gibbs distribution. Because of the Gibbs distribution, Markov random field (MRF) equivalence, this assignment also determines an MRF image model. The energy function is a more convenient and natural mechanism for embodying picture attributes than are the local characteristics of the MRF. For a range of degradation mechanisms, including blurring, nonlinear deformations, and multiplicative or additive noise, the posterior distribution is an MRF with a structure akin to the image model. By the analogy, the posterior distribution defines another (imaginary) physical system. Gradual temperature reduction in the physical system isolates low energy states (``annealing''), or what is the same thing, the most probable states under the Gibbs distribution. The analogous operation under the posterior distribution yields the maximum a posteriori (MAP) estimate of the image given the degraded observations. The result is a highly parallel ``relaxation'' algorithm for MAP estimation. We establish convergence properties of the algorithm and we experiment with some simple pictures, for which good restorations are obtained at low signal-to-noise ratios.
Remote sensing has considerable potential as a source of data for land cover mapping. This potential remains to be fully realised due, in part, to the methods used to extract land cover information from the remotely sensed data. Widely used statistical classifiers provide a poor representation of land cover, make untenable assumptions about the data and convey no information on the quality of individual class allocations. This paper shows that a softened classification, providing information on the strength of membership to all classes for each image pixel, may be derived from a neural network. This information may be used to indicate classification quality on a per-pixel basis. Moreover, a soft or fuzzy classification may be derived to more appropriately represent land cover than the conventional hard classification.
One of the main problems related to unsupervised change detection
methods based on the “difference image” lies in the lack of
efficient automatic techniques for discriminating between changed and
unchanged pixels in the difference image. Such discrimination is usually
performed by using empirical strategies or manual trial-and-error
procedures, which affect both the accuracy and the reliability of the
change-detection process. To overcome such drawbacks, in this paper, the
authors propose two automatic techniques (based on the Bayes theory) for
the analysis of the difference image. One allows an automatic selection
of the decision threshold that minimizes the overall change detection
error probability under the assumption that pixels in the difference
image are independent of one another. The other analyzes the difference
image by considering the spatial-contextual information included in the
neighborhood of each pixel. In particular, an approach based on Markov
Random Fields (MRFs) that exploits interpixel class dependency contexts
is presented. Both proposed techniques require the knowledge of the
statistical distributions of the changed and unchanged pixels in the
difference image. To perform an unsupervised estimation of the
statistical terms that characterize these distributions, they propose an
iterative method based on the Expectation-Maximization (EM) algorithm.
Experimental results confirm the effectiveness of both proposed
techniques
Contenido: Repaso de probabilidades; Modelos de Markov de tiempo discreto; Recurrencia y ergodicidad; Comportamiento durarero; Funciones de Lyapunov y martingalas; Valor propio y cadenas de Markov no homogéneas; Campos de Gibbs y simulación de Monte Carlo; Modelos de Markov de tiempo continuo; Cálculo de Poisson y colas de espera.
A method is proposed for the enhancement of the quality of a classification result by fusing this result with remote sensing images, based on a Markov random field approach. The classification accuracy is estimated by a modified posterior probability, which is used for choosing the optimal classification result. The procedure is applied to a benchmark dataset for discrimination provided by the IEEE Geoscience and Remote Sensing Society Data Fusion Committee, and it shows an excellent performance. The classified result won the competition of the data fusion contest 2001 held by the same committee.
The granular appearance of speckle noise in synthetic aperture radar (SAR) imagery makes it very difficult to visually and automatically interpret SAR data. Therefore, speckle reduction is a prerequisite for many SAR image processing tasks. In this paper, we develop a speckle reduction algorithm by fusing the wavelet Bayesian denoising technique with Markov-random-field-based image regularization. Wavelet coefficients are modeled independently and identically by a two-state Gaussian mixture model, while their spatial dependence is characterized by a Markov random field imposed on the hidden state of Gaussian mixtures. The Expectation-Maximization algorithm is used to estimate hyperparameters and specify the mixture model, and the iterated-conditional-modes method is implemented to optimize the state configuration. The noise-free wavelet coefficients are finally estimated by a shrinkage function based on local weighted averaging of the Bayesian estimator. Experimental results show that the proposed method outperforms standard wavelet denoising techniques in terms of the signal-to-noise ratio and the equivalent-number-of-looks measures in most cases. It also achieves better performance than the refined Lee filter.
This paper addresses the problem of image change detection (ICD) based on Markov random field (MRF) models. MRF has long been recognized as an accurate model to describe a variety of image characteristics. Here, we use the MRF to model both noiseless images obtained from the actual scene and change images (CIs), the sites of which indicate changes between a pair of observed images. The optimum ICD algorithm under the maximum a posteriori (MAP) criterion is developed under this model. Examples are presented for illustration and performance evaluation.
The use of contextual information for modeling the prior
probability mass function has found applications in the classification
of remotely sensed data. With the increasing availability of multisource
remotely sensed data sets, random field models, especially Markov random
fields (MRF), have been found to provide a theoretically robust yet
mathematical tractable way of coding multisource information and of
modeling contextual behavior. It is well known that the performance of a
model is dependent both on its functional form (in this case, the
classification algorithm) and on the accuracy of the estimates of model
parameters. In dealing with multisource data, the determination of
source weighting and MRF model parameters is a difficult issue. The
authors extend the methodology proposed by A. H. Schistad et al. (1996),
by demonstrating that the use of an effective search procedure, the
genetic algorithm, leads to improved parameter estimation and hence
higher classification accuracies
Support Vector Machines for Classification of Multi and Hyperspectral Data
- P Watanachaturaporn
- M K Arora
Watanachaturaporn, P., & Arora, M. K. (2004). Support Vector Machines for Classification of Multi and Hyperspectral Data. In P. K. Varshney, & M. K. Arora (Eds.), Advanced Imaged Processing Techniques for Remotely Sensed Hyperspectral Data. Springer Verlag.