No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Handling large datasets of hyperspectral images: Reducing data size without loss of useful information
... However, their use in the presence of different stressors has yet to be deeply explored (Karimi et al., 2005;Ray, Singh, & Panigrahy, 2010;Strachan et al., 2002). Generation of a hyperspectrogram (HSG) is a data extraction and compression technique based on PCA that manage large numbers of hyperspectral images and retain spatial information (Ferrari, Foca, & Ulrici, 2013). The HSGs are matrix signals that summarise the most relevant information carried by the original hyperspectral images (Ferrari, Foca, Calvini, & Ulrici, 2015). ...
... Thereafter, HSGs represent a valuable approach to reduce the dimension of hyperspectral datasets without losing spectral and spatial information. Although the feasibility of this technique has already been tested in food science (Calvini, Foca, & Ulrici, 2016;Ferrari et al., 2013Ferrari et al., , 2015Xu, Riccioli, & Sun, 2016), it has never been tested on vegetation. ...
... Our goal in this work was to verify the capability of hyperspectral imaging in the VISeNIR spectral range (400e1000 nm) in order to estimate crop variables with multivariate analysis using greenhouse spinach (as crop model) under combined water and nitrogen stresses. As a secondary product, this work compared the performances of two hyperspectral image data extraction methodsdaverage spectra and HSGs calculated per Ferrari et al. (2013) with modifications. ...
This work had the goal to assess the capability of hyperspectral line scan imaging (400–1000 nm) to estimate crop variables in the greenhouse under combined water and nitrogen stress using multivariate data analysis and two data compression methods: canopy average spectra and hyperspectrogram extraction. Hyperspectral images contain far more information than do multispectral ones, which permits discrimination among minute pattern differences in canopy spectral reflectance. A pot greenhouse experiment of eight treatments, from the combination of four nitrogen supply levels and two water supply levels, was designed to test widely varied spinach canopies. Using partial least square regression models, the fresh and dry matter of aboveground biomasses and water and nitrogen contents were estimated from a 76-sample dataset. Both the canopy reflectance-based and hyperspectrogram-based models performed well in estimating variables strictly related to canopy leaf area index (LAI) and geometry, i.e., water content and fresh and dry matters, such that R2 in independent validation reached values of 0.87, 0.65, 0.65, and 0.86, 0.74, 0.72, respectively. Estimation of nitrogen concentration from single leaf spectra hyperspectral images produced a high cross-validation R2 (0.83), as opposed to the poor predictive results produced from canopy scans. This latter result arose from orientation effects due to canopy architecture. Finally, for estimation purposes, image hyperspectrogram compression without spatial information loss produced more encouraging results while considering canopy structure in crop variables than did average canopy spectra.
... Recently, following an approach that was previously developed by some of us for the analysis of datasets of RGB images [10][11][12], we proposed a 1D-DR method, named hyperspectrogram [6,13], to condense both spatial and spectral information of hyperspectral images. Basically, hyperspectrograms are one-dimensional signals built by merging in sequence the frequency distribution curves of pixel-related features obtained from a PCA model calculated separately for each hyperspectral image. ...
... As mentioned above, the idea behind hyperspectrograms is to compress the potentially useful information contained in each hyperspectral image into a signal, by merging together quantities derived by a PCA model calculated on the considered image. A detailed description of the procedure used to calculate SSH is reported in the previously published articles [6,13]; here below, a summary of the procedure used in this work is reported. ...
... The whole dataset of CSH, stored in a file with size equal to 1. 13 It has to be noticed that SSH and CSH were calculated considering a different number of PCs (3 and 5, respectively). Indeed, since the CSH approach is based on a global PCA model, i.e. on a model including all the spectra of all the images of the training set, it accounts for more sources of variance than SSH, which conversely considers only the pixel variability within each single image. ...
Hyperspectral sensors represent a powerful tool for chemical mapping of solid-state samples, since they provide spectral information localized in the image domain in very short times and without the need of sample pretreatment. However, due to the large data size of each hyperspectral image, data dimensionality reduction (DR) is necessary in order to develop hyperspectral sensors for real-time monitoring of large sets of samples with different characteristics. In particular, in this work, we focused on DR methods to convert the three-dimensional data array corresponding to each hyperspectral image into a one-dimensional signal (1D-DR), which retains spectral and/or spatial information. In this way, large datasets of hyperspectral images can be converted into matrices of signals, which in turn can be easily processed using suitable multivariate statistical methods. Obviously, different 1D-DR methods highlight different aspects of the hyperspectral image dataset. Therefore, in order to investigate their advantages and disadvantages, in this work, we compared three different 1D-DR methods: average spectrum (AS), single space hyperspectrogram (SSH) and common space hyperspectrogram (CSH). In particular, we have considered 370 NIR-hyperspectral images of a set of green coffee samples, and the three 1D-DR methods were tested for their effectiveness in sensor fault detection, data structure exploration and sample classification according to coffee variety and to coffee processing method. Principal component analysis and partial least squares-discriminant analysis were used to compare the three separate DR methods. Furthermore, low-level and mid-level data fusion was also employed to test the advantages of using AS, SSH and CSH altogether. [Figure not available: see fulltext.]
... This approach sometimes leads to satisfactory results, especially for homogeneous materials, but it also leads to losing the information related to spatial variability (Ferrari et al., 2014). In order to develop a fast and easy-to-use tool able to analyse large datasets of hyperspectral images while maintaining both spectral-and spatialrelated information, the use of hyperspectrograms was proposed as an approach, which automatically converted each hyperspectral image into a signal and it was proven the effectiveness of the hyperspectrogram-based approach to address a calibration and a defect detection issue in food samples (Ferrari et al., 2013). However, this novel chemometric method has been scarcely applied in the fish industry. ...
... The hyperspectrogram-based approach essentially consists in compressing each hyperspectral image into a signal, named hyperspectrogram (Ferrari et al., 2013). It can be used as a compact set of descriptors and subjected to further multivariate analysis. ...
... As mentioned above, before building models, PCA was conducted to obtain a hyperspectrogram for each sample by calculating on the unfolded hypercube data. During PCA analysis, as suggested by Ferrari et al. (2013), the effects of different column pre-treatments on the resulting models were investigated and summarized in Table 4. Although similar cross-validation and prediction performances were obtained using three pre-treatments (linear detrend, normalization and standard normal variate (SNV)), the best performance in calibration was reported using linear detrend. ...
... In order to deal with datasets composed of a large number of hyperspectral images, our research group recently proposed an alternative approach [6], which is based on the idea of reducing each hyperspectral image into a signal, named hyperspectrogram, built in a way to convey both spatial and spectral information. This procedure is derived from the colourgram approach, which was developed for the elaboration of RGB images [7][8][9]. ...
... According to Ferrari et al. [6], an explorative analysis step by means of PCA was carried out at the pixel level on a few images evaluating several spectral pretreatments, i.e., detrend, first and second derivatives, standard normal variate, and mean centering. These pretreatments were considered both separately and in different combinations, in order to search for the conditions allowing to highlight at best the apple surface defects, as well as to establish the optimal dimensionality of the PCA models. ...
... For a more detailed description of the algorithm used to build hyperspectrograms, the reader is referred to Ferrari et al. [6]. ...
... In the past two decades, a variety of hyperspectral imaging systems have emerged to provide powerful capability for cultural relic conservation. Dvoptic, Headwall, and XI-MEA have developed hyperspectral scanning systems to study paintings [24][25][26]. The VSC6000 system commercialized by Foster+Freeman has been widely used for handwriting analysis [27]. ...
... In the past two decades, a variety of hyperspectral imaging systems have emerged to provide powerful capability for cultural relic conservation. Dvoptic, Headwall, and XIMEA have developed hyperspectral scanning systems to study paintings [24][25][26]. The VSC6000 system commercialized by Foster+Freeman has been widely used for handwriting analysis [27]. ...
In this review, the conservation methods for various types of cultural relics enabled by hyperspectral imaging are summarized, and the hyperspectral cameras and techniques utilized in the process from data acquisition to analyzation are introduced. Hyperspectral imaging is characterized by non-contact detection, broadband, and high resolution, which are of great significance to the non-destructive investigation of cultural relics. However, owing to the wide variety of cultural relics, the utilized equipment and methods vary greatly in the investigations of the associated conservation. Previous studies generally select a single type of cultural relic for conservation. That is, seldom study has focused on the application of hyperspectral techniques to generalized conservation methods that are simultaneously suitable for different types of cultural relics. Hence, some widely used hyperspectral cameras and imaging systems are introduced first. Subsequently, according to the previous investigations, the methods used for image acquisition, image correction, and data dimensionality reduction in hyperspectral techniques are described. Thirdly, a summary of methods in cultural relic conservation based on hyperspectral techniques is presented, which involves pigments, grottoes and murals, and painting and calligraphy. Later, some challenges and potential development prospects in hyperspectral-based methods are discussed for future study. Finally, the conclusions are given.
... Instrumental techniques such as GCMS, e-nose, and hyper-spectral imaging have shown several advantages [6][7][8][9]. But these techniques do suffer from disadvantages such as long sampling time for GCMS, sensor drift response in case of e-nose, and huge data size and processing for hyper-spectral imaging [3,10,11]. Fourier transform infrared spectroscopy (FTIR) offers an alternative solution with a very short sampling time of less than a minute and records the biochemical information of the fruit from its surface. Handheld FTIR has increased its preference over other techniques suitable for a small-scale processing industry as well [12]. ...
... The output layer contained two nodes, one for prediction of TVC and other for Y&M counts. One hidden layer with varying number of neurons (2)(3)(4)(5)(6)(7)(8)(9)(10)(11)(12)(13)(14)(15) to select the best neural network architecture. The best net with the number of neuron in the hidden layer having lowest SEP for validation set was selected. ...
FTIR in combination with chemometric tools was utilised to evaluate the microbial counts for fresh cut jackfruit samples stored at 4 and 10 °C. Predictive models were prepared for total viable counts and yeast & mould counts using partial least square regression (PLS-R) and artificial neural networks (ANN) from FTIR data. Raw FTIR data and its first derivative were exploited for model building. Models built with both ANN and PLS-R using FTIR data demonstrated a high correlation value of R² > 0.85 for 10 °C stored samples. Variable importance projection score obtained from PLS-R models suggested production of acids after utilization of sugars due to microbial activity during storage. Feasibility of utilising FTIR as a rapid non-destructive methodology for estimation of microbial counts for fresh cut jackfruit is demonstrated.
... The background was removed by interactively separating (selecting, excluding and reconstructing) the background pixels from the fruit pixels from a principal component analysis (PCA) based contour plots, applied on hypercubes. Segmented images were exported to a MATLAB (version R2019a, Mathworks, Natick, MA, USA) recognizable format for further processing; converted into MATLAB's dataset objects (DSO) for subsequent use in unfolding 3D hypercubes into 2D data matrices, a technique of dimensionality reduction without loss of information [27]. The exported images were converted into hyperspectrograms for use in subsequent image reconstruction to visualize bruises at latent stage. ...
... In order to show that bruises could be detected at their latent stage, a data visualization was carried out by reconstruction of hyperspectral images in a manner that highlights the bruises. Two methods were used for this purpose, namely hyperspectrograms based image reconstruction [27] and using single wavelength images captured using the Image player tool from MATLAB software. ...
Bruise damage is a very commonly occurring defect in apple fruit which facilitates disease occurrence and spread, leads to fruit deterioration and can greatly contribute to postharvest loss. The detection of bruises at their earliest stage of development can be advantageous for screening purposes. An experiment to induce soft bruises in Golden Delicious apples was conducted by applying impact energy at different levels, which allowed to investigate the detectability of bruises at their latent stage. The existence of bruises that were rather invisible to the naked eye and to a digital camera was proven by reconstruction of hyperspectral images of bruised apples, based on effective wavelengths and data dimensionality reduced hyperspectrograms. Machine learning classifiers, namely ensemble subspace discriminant (ESD), k-nearest neighbors (KNN), support vector machine (SVM) and linear discriminant analysis (LDA) were used to build models for detecting bruises at their latent stage, to study the influence of time after bruise occurrence on detection performance and to model quantitative aspects of bruises (severity), spanning from latent to visible bruises. Over all classifiers, detection models had a higher performance than quantitative ones. Given its highest speed in prediction and high classification performance, SVM was rated most recommendable for detection tasks. However, ESD models had the highest classification accuracy in quantitative (>85%) models and were found to be relatively better suited for such a multiple category classification problem than the rest.
... Hyperspectral imaging can detect and classify the spatial information at high accuracy. However, the high-dimensional feature often causes computational complexity and dimensionality curse (Ferrari et al., 2013;Gu et al., 2017;Sakarya, 2014). In many cases, it is not required to process the information of all spectral bands since many bands are highly correlated. ...
... The high-dimensional data poses many problems, such as computational complexity, dimensionality curse, etc. (Ferrari et al., 2013;Gu et al., 2017). Due to the dimensionality curse and the diminishment of specificity in similarities between points in the hyperspectral image, the complexity of existing methods is exponential with respect to the number of dimensions (Sakarya, 2014). ...
High-dimensional features often cause computational complexity and dimensionality curse. Feature selection and feature extraction are the two mainstream methods for dimensionality reduction. Feature selection but not feature extraction can preserve the critical information and maintain the physical meaning simultaneously. Herein, we proposed a dimensionality reduction method based on rough set theory (DRM-RST) for feature selection. We defined the hyperspectral image as a decision system, extracted the features as decision attributes, and selected the effective features based on information entropy. We used the Washington D.C. Mall dataset and New York dataset to evaluate the performance of DRM-RST on dimensionality reduction. Compared with full band classification, 184 or 185 redundant bands were removed in DRM-RST, respectively. DRM-RST achieved similar accuracy (overall accuracy >94%) by SVM classifier and reduced computing time by about 85%. We further compared the dimensionality reduction efficiency of DRM-RST against other popular methods, including ReliefF, Sequential Backward Elimination (SBE) and Information Gain (IG). The Producer’s accuracy (PA) and User’s accuracy (UA) of DRM-RST was greater than that of ReliefF and IG. DRM-RST showed greater stability of accuracy than SBE in dimensionality reduction when using for different datasets. Collectively, this study provides a new method for dimensionality reduction that can reduce computational complexity and alleviate dimensionality curse.
... CNNs and RF models have been shown to handle large datasets with high accuracy, but further research into lightweight models optimized for real-time analysis is needed. • Data compression techniques: Improvements in data compression technologies can reduce the size of HSI datasets without losing important spectral information (Dua et al., 2020;Ferrari et al., 2013). Algorithms such as principal component analysis (PCA) or wavelet transform could be applied to reduce data dimensionality, making storage and real-time analysis more feasible (Elgargni et al., 2015). ...
Ensuring global food security in the face of growing population, climate change, and resource limitations is a critical challenge. Hyperspectral imaging (HSI), particularly when combined with drone technology, offers innovative solutions to enhance agricultural productivity and food quality by providing detailed, real-time data on crop health, disease detection, water and nutrient management, and post-harvest quality control. This review highlights the applications of drone-based HSI in precision agriculture, where it enables early detection of crop stress, accurate yield prediction, and soil health assessment. In post-harvest management, HSI is utilized to monitor food freshness and ripeness and detect potential contaminants, improving food safety and reducing waste. While the benefits of HSI are significant, challenges such as managing large volumes of data, translating spectral information into actionable insights, and ensuring cost-effective access for smallholder farmers remain barriers to its widespread adoption. Looking forward, future directions include advancements in miniaturized sensors, integration with Internet of Things (IoT) devices and satellite data for comprehensive agricultural monitoring, and expanding HSI applications to precision animal sciences. Collaboration among researchers, policymakers, and industry will be crucial to scaling the impact of HSI on global food systems, ensuring sustainable and equitable access to technology.
... Other widely used statistical techniques that lead to the compression of hyperspectral data includes Principal Component Analysis (PCA) [13][14][15] and Independent Component Analysis (ICA). 14, 16 PCA transforms the original set of variables into a new set of principal components (uncorrelated variables or orthogonal), ordered so that the first few retain most of the variation present in all of the original variables. ...
Purpose
Hyperspectral imaging (HSI) collects detailed spectral information across hundreds of narrow bands, providing valuable datasets for applications such as medical diagnostics. However, the large size of HSI datasets, often exceeding several gigabytes, creates significant challenges in data transmission, storage, and processing. This study aims to develop a wavelet-based compression method that addresses these challenges while preserving the integrity and quality of spectral information.
Approach
The proposed method applies wavelet transforms to the spectral dimension of hyperspectral data in three steps: 1) wavelet transformation for dimensionality reduction, 2) spectral cropping to eliminate low-intensity bands, and 3) scale matching to maintain accurate wavelength mapping. Daubechies wavelets were used to achieve up to 32x compression while ensuring spectral fidelity and spatial feature retention.
Results
The wavelet-based method achieved up to 32x compression, corresponding to a 96.88% reduction in data size without significant loss of important data. Unlike PCA and ICA, the method preserved the original wavelength scale, enabling straightforward spectral interpretation. Additionally, the compressed data exhibited minimal loss in spatial features, with improvements in contrast and noise reduction compared to spectral binning.
Conclusions
This study demonstrates that wavelet-based compression is an effective solution for managing large HSI datasets in medical imaging. The method preserves critical spectral and spatial information, and therefore facilitates efficient data storage and processing, providing the way for practical integration of HSI technology in clinical applications.
... Several studies have demonstrated the potential for improving model performance by decreasing dataset complexity. For example, by leveraging techniques that reduce redundancy or compress the data structure, models can achieve faster convergence and higher accuracy, as reported in recent literature [41]. Thus, while we optimized our approach using generalized Dirichlet distributions and geometric transformations, reducing the inherent complexity of the dataset itself could further contribute to enhancing the model's accuracy and computational efficiency. ...
We address the problem of anomaly detection in data by learning a normality score function through the use of data transformations. Applying transformations to a dataset is essential for enhancing its representation and revealing underlying patterns. First, we propose geometric transformations for image data. The core idea of our approach is to train a multi-class deep classifier to distinguish between various geometric transformations. At test time, we construct the normality score function by approximating the softmax output predictions vector using generalized forms of Dirichlet distributions, including the generalized Dirichlet (GD), scaled Dirichlet (SD), shifted scaled Dirichlet (SSD), and Beta-Liouville (BL) distributions. These generalized forms of the Dirichlet distribution are more robust in real-world applications compared to the standard Dirichlet distribution. They offer a more flexible covariance structure, making them suitable for approximating both symmetric and asymmetric distributions. For parameter estimation, we use the maximum likelihood method based on the transformed forms of the original data. In the second step, we extend our approach to non-image data by selecting appropriate transformations. This transformation procedure involves building several neural networks, training them on the original data to obtain its transformed form, and then passing the transformed data through an auto-encoder. Experiments conducted on both image and non-image data demonstrate the effectiveness of our proposed strategy. The results show that our anomaly detection models, based on generalized Dirichlet distributions, outperform baseline techniques and achieve high Area Under the Receiver Operating Characteristic (AUROC) scores.
... This toy example makes evident that hyperspectral images are high dimensional data arrays, and their analysis involves issues related to data handling and storage [33]. These problems become even more relevant considering that for practical applications it is necessary to acquire a large number of images and analyse them altogether to compare samples in different images or to evaluate changes over time of sample composition [34,35]. ...
Sparse-based models are a powerful tool for data compression, variable reduction, and model complexity reduction. Nevertheless, their major issue is the high computational time needed in large matrices. This manuscript proposes, for the first time, to couple randomised decomposition as a first step before sparsity calculations, followed by a projection of the full data onto a reduced-sparse set of loadings that will drastically reduce the time needed for calculations and built models that are equally reliable as their sparse-based homologous. While this new approach might be valid for several scenarios (exploration, regression and classification), we will focus on exploration methods (like Principal Component Analysis – PCA) applied to large datasets of hyperspectral images. Two datasets of different complexity have been tested, and the benefits of the coupled randomisation and sparse PCA (rsPCA) are extensively studied.
... This method captures the necessary and most important variance for each sample. This approach is similar to applying convex hull [22][23] or other data reduction strategies [24], subsequent to PCA. The 15 selected spectra of each sample, for both the VNIR and SWIR data sets, were concatenated into two data blocks resulting in a total of 5700 spectra each. ...
... This technique allows a spatially resolved sample analysis by detecting the different bonding vibrations of the tissue molecules. Commonly, FTIR spectroscopy is also combined with multivariate data analysis (MVA) to systematically reduce large data sets and extract the most relevant tissue-related information [14][15][16][17]. Many studies were performed using FTIR spectroscopy or imaging coupled with MVA in tumor diagnostics [18][19][20][21]. ...
Due to the wide variety of benign and malignant salivary gland tumors, classification and malignant behavior determination based on histomorphological criteria can be difficult and sometimes impossible. Spectroscopical procedures can acquire molecular biological information without destroying the tissue within the measurement processes. Since several tissue preparation procedures exist, our study investigated the impact of these preparations on the chemical composition of healthy and tumorous salivary gland tissue by Fourier-transform infrared (FTIR) microspectroscopy. Sequential tissue cross-sections were prepared from native, formalin-fixed and formalin-fixed paraffin-embedded (FFPE) tissue and analyzed. The FFPE cross-sections were dewaxed and remeasured. By using principal component analysis (PCA) combined with a discriminant analysis (DA), robust models for the distinction of sample preparations were built individually for each parotid tissue type. As a result, the PCA-DA model evaluation showed a high similarity between native and formalin-fixed tissues based on their chemical composition. Thus, formalin-fixed tissues are highly representative of the native samples and facilitate a transfer from scientific laboratory analysis into the clinical routine due to their robust nature. Furthermore, the dewaxing of the cross-sections entails the loss of molecular information. Our study successfully demonstrated how FTIR microspectroscopy can be used as a powerful tool within existing clinical workflows.
... Issues common to all hyperspectral imager types are the significant computing power required and the large file sizes of the data cubes, especially in applications involving larger fields of view. Attempts to address these issues have included the application of compressive sensing [17][18][19], deep neural networks [20], and methods centered around principal component analysis (PCA) [21]. Each of these solutions has its own limitations in terms of heavy computational requirements and large file sizes for data cube analysis. ...
Relative to standard red/green/blue (RGB) imaging systems, hyperspectral imaging systems offer superior capabilities but tend to be expensive and complex, requiring either a mechanically complex push-broom line scanning method, a tunable filter, or a large set of light emitting diodes (LEDs) to collect images in multiple wavelengths. This paper proposes a new methodology to support the design of a hypothesized system that uses three imaging modes—fluorescence, visible/near-infrared (VNIR) reflectance, and shortwave infrared (SWIR) reflectance—to capture narrow-band spectral data at only three to seven narrow wavelengths. Simulated annealing is applied to identify the optimal wavelengths for sparse spectral measurement with a cost function based on the accuracy provided by a weighted k-nearest neighbors (WKNN) classifier, a common and relatively robust machine learning classifier. Two separate classification approaches are presented, the first using a multi-layer perceptron (MLP) artificial neural network trained on sparse data from the three individual spectra and the second using a fusion of the data from all three spectra. The results are compared with those from four alternative classifiers based on common machine learning algorithms. To validate the proposed methodology, reflectance and fluorescence spectra in these three spectroscopic modes were collected from fish fillets and used to classify the fillets by species. Accuracies determined from the two classification approaches are compared with benchmark values derived by training the classifiers with the full resolution spectral data. The results of the single-layer classification study show accuracies ranging from ~68% for SWIR reflectance to ~90% for fluorescence with just seven wavelengths. The results of the fusion classification study show accuracies of about 95% with seven wavelengths and more than 90% even with just three wavelengths. Reducing the number of required wavelengths facilitates the creation of rapid and cost-effective spectral imaging systems that can be used for widespread analysis in food monitoring/food fraud, agricultural, and biomedical applications.
... In other words, only 16 readings enable the reliable recovery of the original spectra in the broad spectral range of λ = 450−700 nm. Considering a large size of typical hyperspectral image data, 57 another potential application of light localization-based spectral information processing is spectral data compression; if compressive sampling and recovery are used as an encoder and a decoder, 58 the salient spectral information can be retained using significantly less storage (m = 16; 128 bytes with double precision) than Nyquist (1 nm sampling for a spectral resolution of 2.1 nm) sampled data (q = 250; 2000 bytes with double precision) with an average root mean square error of 1.0 × 10 −3 in the blood hemoglobin testing samples. ...
Information recovery from incomplete measurements, typically performed by a numerical means, is beneficial in a variety of classical and quantum signal processing. Random and sparse sampling with nanophotonic and light scattering approaches has received attention to overcome the hardware limitations of conventional spectrometers and hyperspectral imagers but requires high-precision nanofabrications and bulky media. We report a simple spectral information processing scheme in which light transport through an Anderson-localized medium serves as an entropy source for compressive sampling directly in the frequency domain. As implied by the “lustrous” reflection originating from the exquisite multilayered nanostructures, a pearl (or mother-of-pearl) allows us to exploit the spatial and spectral intensity fluctuations originating from strong light localization for extracting salient spectral information with a compact and thin form factor. Pearl-inspired light localization in low-dimensional structures can offer an alternative of spectral information processing by hybridizing digital and physical properties at a material level.
... For example, paperboard is particularly difficult to differentiate from feed material, making its detection complicated and time consuming. In order to overcome this issue, the stereomicroscope could be coupled with more advanced imaging systems capable of detecting light also beyond the visible spectral region, including for example the near infrared spectral region (Gowen et al. 2011;Ferrari et al. 2013;Dale et al. 2013;Ulrici et al. 2013;Amigo et al. 2015;Calvini et al. 2016Calvini et al. , 2018Vermeulen et al. 2017). The colour features of the samples derived from RGB images could be combined with spectral features derived from multispectral or hyperspectral images, thus leading to a more comprehensive characterisation of the differences between former food matrices and residues of packaging materials. ...
From a circular economy perspective, feeding livestock with food leftovers or former foodstuff products (FFPs) could be an effective option aimed at exploiting food leftover resources and reducing food losses. FFPs are valuable energy sources, characterised by a beneficial starch/sugar content, and also fats. However, besides these nutritional aspects, safety is a key concern given that FFPs are generally derived from packaged food. Packaging materials, such as plastics and paper, are not accepted as a feed ingredient which means that residues should be rigorously avoided. A sensitive and objective detection method is thus essential for an accurate risk evaluation throughout the former food production chain. To this end, former food samples were collected in processing plants of two different European countries and subjected to multivariate analysis of red, green, and blue (RGB) microscopic images, in order to evaluate the possible application of this non-destructive technique for the rapid detection of residual particles from packaging materials. Multivariate Image Analysis (MIA) was performed on single images at the pixel level, which essentially consisted in an exploratory analysis of the image data by means of Principal Component Analysis, which highlighted the differences between packaging and foodstuff particles, based on their colour. The whole dataset of images was then analysed by means of a multivariate data dimensionality reduction method known as the colourgrams approach, which identified clusters of images sharing similar features and also highlighted outlier images due to the presence of packaging particles. The results obtained in this feasibility study demonstrated that MIA is a promising tool for a rapid automated method for detecting particles of packaging materials in FFPs. ARTICLE HISTORY
... An issue common to all hyperspectral imager types is the significant computing power required and the large file sizes of the data cubes, especially in applications involving larger fields of view. Attempts to address these issues have included the application of compressive sensing [11], [12], [13] , deep neural networks [14] , and methods centered around principal component analysis (PCA) [15] . Each of these solutions has its own limitations in terms of heavy computational requirements and large file sizes for data cube analysis. ...
... Hyperspectral imaging technology, in which the spectral resolution is generally less than 10 nm, is a comparatively safe (for relics) and nondestructive method for the detection and restoration of relics. [1][2][3][4][5] Compared to portable X-ray fluorescence spectroscopy, Raman spectroscopy, and various other methods for analyzing microdamage, hyperspectral imaging can be used to safely and rapidly obtain large-scale images and reflection spectra. In hyperspectral images, different substances show reactions that vary by band, facilitating the analysis thereof, and thus revealing hidden information pertaining to cultural relics. ...
Hyperspectral technology is a nondestructive, fast, and reliable method for the detection and restoration of relics. Most of the band characteristics of mineral pigment are concentrated between 2200 and 2400 nm, and these data are expensive to obtain (the required imaging sensor is expensive). We are pursuing a hyperspectral index mean that can effectively distinguish pigments in shorter band ranges to achieve high application value that is much less expensive. In this study, based on the spectral features of azurite at 400-1500 nm, we created an azurite normalized difference spectral index (ANDSI) through feature band selection, derivation of characteristic formulae, and discrimination analysis. Reflectivity bands at 458, 806, and 1373 nm were selected to build the ANDSI. Azurite was compared with 25 other common pigments and it was found that the discrimination values between azurite and the other pigments exceeded 0.88 (where values >0.5 indicate discriminable pigments), demonstrating that the ANDSI is suitable for detecting azurite.
... Furthermore, since the colourgrams are obtained by merging in sequence the frequency distribution curves of several colour parameters, it is possible to identify colour features of interest and visualise them back at the pixel level in the original image domain. A similar approach has also been successfully proposed for the analysis of hyperspectral images (Calvini, Foca, & Ulrici, 2016;Ferrari, Foca, & Ulrici, 2013). In the specific case under investigation, the acquired RGB images of the raw ham samples were converted into colourgrams and the signals were visually inspected in order to identify the colour parameter leading to the optimal segmentation of the ROI. ...
Veins in pork thigh carcass are directly related to the quality of dry-cured ham, and consequently to its market value. Some veining defects over the surface of raw ham are easily detected by humans and precisely assessed by a specialist. However, the automatic evaluation of raw ham quality by image analysis poses some challenges to the traditional Computer Vision Systems (CVS), many of them grounded on the complex image pattern related to each defect level. To improve the CVS classification performance without overburdening feature extraction, as well as the common machine learning modelling, we propose Dual Stage Image Analysis (DSIA). DSIA is an additional step in a CVS, that was built in two stages based on the “divide and conquer” strategy. The first stage consists of splitting the region of interest into sub-regions to predict the presence of veining. In the second stage, the algorithm computes the number of veining sub-regions to assess the final defect level classification. A total of 194 raw ham samples were used to evaluate the DSIA performance in the experiments. Support Vector Machine and Random Forest algorithms were compared for inducing the classification model using 92 image features. Random Forest model was the best, capable of predicting defect level with 88.10% accuracy using DSIA. Without DSIA, the CVS with RF achieved an accuracy of 63.10%.
... To this aim, a data dimensionality reduction method has been proposed, which consists in converting each hyperspectral image of the dataset into a one-dimensional signal, named hyperspectrogram, obtained by merging in sequence the frequency distribution curves of quantities derived from a PCA model calculated on the images (Calvini, Foca, & Ulrici, 2016;Corti, Gallina, Cavalli, & Cabassi, 2017;Ferrari et al., 2013Ferrari et al., , 2015Xu, Riccioli, & Sun, 2016). In this manner, each hyperspectrogram summarizes the relevant information contained in the corresponding hyperspectral image and a large dataset of hyperspectral images is converted into a matrix of signals, which in turn can be analysed by means of common chemometric methods. ...
Parmigiano Reggiano (P-R) is one of the most important Italian food products labelled with Protected Designation of Origin (PDO). The PDO denomination is applied also to grated P-R cheese products meeting the requirements regulated by the Specifications of Parmigiano Reggiano Cheese. Different quality parameters are monitored, including the percentage of rind, which is edible and should not exceed the limit of 18% (w/w). The present study aims at evaluating the possibility of using near infrared hyperspectral imaging (NIR-HSI) to quantify the rind percentage in grated Parmigiano Reggiano cheese samples in a fast and non-destructive manner. Indeed, NIR-HSI allows the simultaneous acquisition of both spatial and spectral information from a sample, which is more suitable than classical single-point spectroscopy for the analysis of heterogeneous samples like grated cheese. Hyperspectral images of grated P-R cheese samples containing increasing levels of rind were acquired in the 900–1700 nm spectral range. Each hyperspectral image was firstly converted into a one-dimensional signal, named hyperspectrogram, which codifies the relevant information contained in the image. Then, the matrix of hyperspectrograms was used to calculate a calibration model for the prediction of the rind percentage using Partial Least Squares (PLS) regression. The calibration model was validated considering two external test sets of samples, confirming the effectiveness of the proposed approach.
... The effectiveness of this approach induced us to adapt it also to the analysis of large datasets of hyperspectral images, leading to the hyperspectrograms approach, which allowed to obtain satisfactory results in various applications [19,[29][30][31][32][33]. ...
Colourgrams GUI is a graphical user-friendly interface developed in order to facilitate the analysis of large datasets of RGB images through the colourgrams approach. Briefly, the colourgrams approach consists in converting a dataset of RGB images into a matrix of one-dimensional signals, the colourgrams, each one codifying the colour content of the corresponding original image. This matrix of signals can be in turn analysed by means of common multivariate statistical methods, such as Principal Component Analysis (PCA) for exploratory analysis of the image dataset, or Partial Least Squares (PLS) regression for the quantification of colour-related properties of interest.
Colourgrams GUI allows to easily convert the dataset of RGB images into the colourgrams matrix, to interactively visualize the signals coloured according to qualitative and/or quantitative properties of the corresponding samples and to visualize the colour features corresponding to selected colourgram regions into the image domain. In addition, the software also allows to analyse the colourgrams matrix by means of PCA and PLS.
... To our knowledge, no other studies are published on the utilization of HSI for asbestos and ACMs classification inside C&DW waste at laboratory scale. Despite the many advantages provided by this technique, a wider diffusion of HSI is hampered by the high amount of data that can be collected in very short times: it is thus necessary a to utilize robust and reliable statistical approaches to manage the data [23,24]. The proposed strategy, based on the combined use of HSI and chemometric techniques, can represent a valid and efficient innovative analytical approach that can support the currently adopted techniques for asbestos recognition [25,26]. ...
Asbestos-Containing Materials (ACMs) are hazardous and prohibited to be sold or used as recycled materials. In the past, asbestos was widely used, together with cement, to produce "asbestos cement-based" products. During the recycling process of Construction and Demolition waste (C&DW), ACM must be collected and deposited separately from other wastes. One of the main aims of the recycling strategies applied to C&DW was thus to identify and separate ACM from C&DW (e.g., concrete and brick). However, to obtain a correct recovery of C&DW materials, control methodologies are necessary to evaluate the quality and the presence of harmful materials, such as ACM. HyperSpectral Imaging (HSI)-based sensing devices allow performing the full detection of materials constituting demolition waste. ACMs are, in fact, characterized by a spectral response that nakes them is different from the "simple" matrix of the material/s not embedding asbestos. The described HSI quality control approach is based on the utilization of a platform working in the shortwave infrared range (1000-2500 nm). The acquired hyperspectral images were analyzed by applying different chemometric methods: Principal Component Analysis for data exploration and hierarchical Partial Least-Square-Discriminant Analysis (PLS-DA) to build classification models. Following this approach, it was possible to set up a repeatable, reliable and efficient technique able to detect ACM presence inside a C&DW flow stream. Results showed that it is possible to discriminate and identify ACM inside C&DW. The recognition is potentially automatic, non-destructive and does not need any contact with the investigated products.
... Hyperspectral imaging (HSI) refers to a technique that can provide both spatial and spectral information by integrating two classical optical sensing technologies of imaging and spectroscopy into one system (Ferrari et al., 2013;Ravikanth et al., 2017). It was mainly applied on remote sensing in the early 70s. ...
This study aims to investigate the potential of an original polarized hyperspectral imaging (HSI) setup in the spectral domain of 400-1000 nm for sunflower leaves in real-world. Dataset 1 includes hypercubes of sunflower leaves in two varieties with different life growth stages, while Dataset 2 is comprised of healthy and contaminated sunflower leaves suffering from powdery mildew (PM) and/or septoria leaf spot (SLS). Cross polarised (R_(⊥ )), parallel polarised (R_(||)) reflectance signals, R_BS (R_(||)+R_(⊥ )) and R_SS (R_(||)-R_(⊥ )) spectra were obtained and used to develop partial least squares-discriminant analysis (PLS-DA) models. Surface information played an important role in separating two varieties of leaves due to the fact that the best model performance was achieved by using R_SS mean spectra, while both surface and subsurface were equally important in classifying leaves between two major growth stages because model of R_BS mean spectra outperformed other models. The best classification model for disease detection was achieved by using pixel R_(⊥ ) spectra with the correct classification rate (CCR) of 0.963 for both cross validation and prediction, meaning that subsurface spectral features were the most important to detect infected leaves. The resulting classification maps were also displayed to visualize the distribution of the infected regions on the leaf samples. The overall results obtained in this research showed that the developed polarized-HSI system coupled with multivariate analysis has considerable promise in real-world applications.
... Two data compression methods were applied: canopy average spectra and hyperspectrogram extraction. The latter is a data extraction and compression technique based on a PCA that manages large numbers of hyperspectral images while retaining spatial information (46). Multivariate data analysis models performed well in estimating the variables. ...
Plant disease detection represents a tremendous challenge for research and practical applications. Visual assessment by human raters is time-consuming, expensive, and error prone. Disease rating and plant protection need new and innovative techniques to address forthcoming challenges and trends in agricultural production that require more precision than ever before. Within this context, hyperspectral sensors and imaging techniques—intrinsically tied to efficient data analysis approaches—have shown an enormous potential to provide new insights into plant-pathogen interactions and for the detection of plant diseases. This article provides an overview of hyperspectral sensors and imaging technologies for assessing compatible and incompatible plant-pathogen interactions. Within the progress of digital technologies, the vision, which is increasingly discussed in the society and industry, includes smart and intuitive solutions for assessing plant features in plant phenotyping or for making decisions on plant protection measures in the context of precision agriculture.
... In contrast to the natural image, the multi-channel image contains higher dimension, requiring (1) a particular method of reducing the data size to reduce the computation cost; and (2) a specific architecture of deep learning model to handle the correlation among channels. For the former problem, Ferrari et al. proposed a method for compression of hyperspectral images based on principal component analysis (PCA) without losing spectral and spatial information [35]. However, little literature is found about the latter one. ...
Deep learning has become a widely used powerful tool in many research fields, although not much so yet in agriculture technologies. In this work, two deep convolutional neural networks (CNN), viz. Residual Network (ResNet) and its improved version named ResNeXt, are used to detect internal mechanical damage of blueberries using hyperspectral transmittance data. The original structure and size of hypercubes are adapted for the deep CNN training. To ensure that the models are applicable to hypercube, we adjust the number of filters in the convolutional layers. Moreover, a total of 5 traditional machine learning algorithms, viz. Sequential Minimal Optimization (SMO), Linear Regression (LR), Random Forest (RF), Bagging and Multilayer Perceptron (MLP), are performed as the comparison experiments. In terms of model assessment, k-fold cross validation is used to indicate that the model performance does not vary with the different combination of dataset. In real-world application, selling damaged berries will lead to greater interest loss than discarding the sound ones. Thus, precision, recall, and F1-score are also used as the evaluation indicators alongside accuracy to quantify the false positive rate. The first three indicators are seldom used by investigators in the agricultural engineering domain. Furthermore, ROC curves and Precision-Recall curves are plotted to visualize the performance of classifiers. The fine-tuned ResNet/ResNeXt achieve average accuracy and F1-score of 0.8844/0.8784 and 0.8952/0.8905, respectively. Classifiers SMO/ LR/RF/Bagging/MLP obtain average accuracy and F1-score of 0.8082/0.7606/0.7314/0.7113/0.7827 and 0.8268/0.7796/0.7529/0.7339/0.7971, respectively. Two deep learning models achieve better classification performance than the traditional machine learning methods. Classification for each testing sample only takes 5.2 ms and 6.5 ms respectively for ResNet and ResNeXt, indicating that the deep learning framework has great potential for online fruit sorting. The results of this study demonstrate the potential of deep CNN application on analyzing the internal mechanical damage of fruit.
... HSI involves considerably higher requirements than NIRS in terms of data processing, due to the larger dataset. Recent reviews have been published on these aspects (9,10). ...
Hyperspectral imaging (HSI) combines spectroscopy and imaging, providing information about the chemical properties of a material and their spatial distribution. It represents an advance of traditional Near-Infrared (NIR) spectroscopy. The present work reviews the most recent applications of NIR spectroscopy for cereal grain evaluation, then focused on the use of HSI in this field. The progress of research from ground material to whole grains and single kernels is detailed. The potential of NIR-based methods to predict protein content, sprout damage and α-amylase activity in wheat and barley is shown, in addition to assessment of quality parameters in other cereals such as rice, maize and oats, and the estimation of fungal infection. This analytical technique also offers the possibility to rapidly classify grains based on properties such as variety, geographical origin, kernel hardness, etc. Further applications of HSI are expected in the near future, for its potential for rapid single-kernel analysis.
... To this aim, it is therefore necessary to define a parameter which relates the sparsity level to the desired performance of the sPCA model.Frequency distribution curves calculated from the score vectors of different sPCA models can be used in order to solve this issue. As a matter of fact, successful approaches based on frequency distribution curves of quantities derived from PCA models have been already reported in the literature for the characterization of whole hyperspectral images[47,48] and for the discrimination of objects within the image scene[49].In this chapter, we describe the use of frequency distribution curves of the sPCA scores in order to optimize the c value and the number of sPCs. In particular, the applications described in the following sections are related to two different issues that generally may occur when dealing with HSI data: the discrimination between two groups of homogeneous samples and the identification of outlier regions within the image scene (i.e., of outlier pixels). ...
One of the main issues of hyperspectral imaging data is to unravel the relevant, yet overlapped, huge amount of information contained in the spatial and spectral dimensions. When dealing with the application of multivariate models in such high-dimensional data, sparsity can improve the interpretability and the performance of the model. In this chapter, we will introduce the exploration of hyperspectral images using a sparse version of the well-known principal component analysis method to demonstrate how the derived models can reveal very useful spectral zones. In particular, we will present two practical applications related to different issues: the separation among groups of homogeneous samples and the identification of outlier pixels in the spatial domain. For both case studies, guidance to the identification of the proper level of sparsity will be provided and, furthermore, we will show how sparsity can improve the chemical interpretation of the results.
... The two datasets of segmented images, one for the Petri dishes and one for the ham samples, were converted into sets of onedimensional signals, called hyperspectrograms, by means of a recently developed algorithm [28], which is derived from the colourgram approach, previously developed for the processing of RGB images [29][30][31]. ...
... Therefore, data reduction is frequently needed in order to preserve only the useful information contained in high-dimensional data [5,6]. ...
... Generally the first PC accounts for the maximum variance of the data, and the second accounts for the maximum of the residual variance, and so on. Thus, in order to have an estimate of the number of PCs potentially bringing useful information, a preliminary evaluation by PCA on a restricted number of representative images can be very useful (Ferrari et al., 2013). After the PCA process on the cleaned image and the tentative interaction analysis including PC score image, score plot and variance percentage of PCs, only the first 8 PCs were adopted for follow-up information extraction and classification process, since noise has become a dominant component of the rest PCs which can be seen clearly from corresponding PC score images. ...
The feasibility of detecting Aflatoxin B1 (AFB1) in single maize kernel inoculated with Aspergillus flavus conidia in the field, as well as its spatial distribution in the kernels, was assessed using near-infrared hyperspectral imaging (HSI) technique. Firstly, an image mask was applied to a pixel-based image mosaic to remove background and shading. Secondly, bad lines in spectra imaging caused by inherent defects of Mercury Cadmium Telluride (MCT) detector were removed through an interactive analysis based on principal component analysis (PCA). Then a PCA procedure was carried out again on the cleaned image, key wavelengths such as 1729 and 2344 nm were shown clearly from the loading line plot of the seventh principal component (PC7). And the pixel of AFB1 extracted from the 5-dimensional scatter plot space formed by five principal components (PCs) from PC4 to PC8 (especially PC7 and PC5) were taken as the input of the spectral angle mapper (SAM) classifier, accuracies of the three varieties of kernels reached 96.15%, 80%, and 82.61% respectively if kernels containing either high (⩾100 ppb) or low (<10 ppb) levels of aflatoxin. A slightly better test result could be got if the kernels placed with different germ orientation. Finally, the repeatability was verified using the fourth variety of kernels.
Harmful algae blooms (HABs) pose critical threats to aquatic ecosystems and human economies, driven by their rapid proliferation, oxygen depletion capacity, toxin release, and biodiversity impacts. These blooms, increasingly exacerbated by climate change, compromise water quality in both marine and freshwater ecosystems, significantly affecting marine life and coastal economies based on fishing and tourism while also posing serious risks to inland water bodies. This article examines the role of hyperspectral imaging (HSI) in monitoring HABs. HSI, with its superior spectral resolution, enables the precise classification and mapping of diverse algae species, emerging as a pivotal tool in environmental surveillance. An array of HSI techniques, algorithms, and deployment platforms are evaluated, analyzing their efficacy across varied geographical contexts. Notably, hyperspectral sensor-based studies achieved up to 90% classification accuracy, with regression-based chlorophyll-a (Chl-a) estimations frequently reaching coefficients of determination (R2) above 0.80. These quantitative findings underscore the potential of HSI for robust HAB diagnostics and early warning systems. Furthermore, we explore the current limitations and future potential of HSI in HAB management, highlighting its strategic importance in addressing the growing environmental and economic challenges posed by HABs. This paper seeks to provide a comprehensive insight into HSI’s capabilities, fostering its integration in global strategies against HAB proliferation.
The present study is focused on the evaluation of the effect of grater type and fat content of the pulp on the spectral response obtained by near infrared hyperspectral imaging (NIR-HSI), when this technique is used to determine the rind percentage in Parmigiano Reggiano (P-R) cheese.
To this aim, grated P-R cheese samples were prepared considering all the possible combinations between three levels of rind amount (8%, 18% and 28%), two levels of fat content of the pulp and two different grater types, and the corresponding hyperspectral images were acquired in the 900–1700 nm spectral range. In a first step, the average spectrum (AS) was calculated from each hyperspectral image, and the corresponding dataset was analysed by means of Analysis of Variance Simultaneous Component Analysis (ASCA) to assess the effect of the three considered factors and their two-way interactions on the spectral response. Then, the hyperspectral images were converted into Common Space Hyperspectrograms (CSH), which are signals obtained by merging in sequence the frequency distribution curves of quantities calculated from a Principal Component Analysis (PCA) model common to the whole hyperspectral image dataset. ASCA was also applied to the CSH dataset, in order to evaluate the effect of the considered factors on this kind of signals.
Generally, all the three factors resulted to have a significant effect, but with a different extent according to the method used to analyse the hyperspectral images. Indeed, while fat content of the pulp and rind percentage showed a comparable effect on the spectral response of AS dataset, in the case of CSH signals rind percentage had a greater effect compared to the other main factors. However, CSH were also more sensitive to differences ascribable to the natural variability between diverse Parmigiano Reggiano cheese samples.
The special nature of hyperspectral imaging (HSI) data requires special image analysis treatments using mathematical, statistical, and software programming approaches. These operations are crucial in building an automatic computer-integrated HSI system qualified for nondestructive assessment of various food quality traits. The theory, fundamentals, and principles of such a system and all accompanying methods associated with the development of robust image processing algorithms of hyperspectral images are explored and reported in this chapter. The quality of the acquired hyperspectral images, the way of extracting spectral fingerprints, and methods of data modeling have substantial effects on the outcomes of the analyses and processing. Fundamental image analysis operations experienced on hyperspectral images during food quality evaluation processes are the cornerstone of this technique. The explored methodologies will have positive impacts not only for food engineers and scientists but also for the food industry willing to adopt this technology in their premises. The strategy applied for image processing for analyzing and visualizing the final results is extremely important to identify the proper decision in detection, classification, quantification, and/or prediction processes. The applications of HSI systems in different sorts of agrifood products were reported with specific examples to demonstrate the potential of such systems in a wide range of analytical tasks. At the end of this chapter, the reader can realize the great capabilities of HSI systems as a novel emerging technique for noninvasive estimation of quality parameters, which proofs why this technology received great acceptance from scientific communities and gained a rapid interest from researchers and food industries. Therefore the state of the art for HSI is expected to gain more and more applications in food analysis and characterization.
An approach is proposed and illustrated for the joint selection of essential samples and essential variables of a data matrix in the frame of spectral unmixing. These essential features carry the signals required to linearly recover all the information available in the rows and columns of a data set. Working with hyperspectral images, this approach translates into the selection of essential spectral pixels (ESPs) and essential spatial variables (ESVs). This results in a highly-reduced data set, the benefits of which can be minimized computational effort, meticulous data mining, easier model building as well as better problem understanding or interpretation. Working with both simulated and real data, we show that (i) reduction rates of over 99% can be typically obtained, (ii) multivariate curve resolution – alternating least squares (MCR-ALS) can be easily applied on the reduced data sets and (iii) the full distribution maps and spectral profiles can be readily obtained from the reduced profiles and the reduced data sets (without using the full data matrix).
This article describes the development of a fast and inexpensive method based on digital image analysis for the automated quantification of the percentage of defective maize (%DM). Defective kernels tend to foster high levels of mycotoxins like Deoxynivalenol (DON), which represents a risk for the health of humans and of farm animals. In this work, 332 RGB images of 83 mixtures containing different amounts of defective maize kernels were acquired using a digital camera. The mixtures were also analysed with a commercial ELISA test kit to determine their concentration of DON, that resulted highly correlated with the amount of defective kernels. Each image was then converted into a signal, named colourgram, which codifies its colour-related information content. The colourgrams were firstly explored using Principal Component Analysis. Then, calibration models of the %DM values were developed using Partial Least Squares (PLS) and interval-PLS. The best interval-PLS model allowed to predict the %DM values of external test set samples with a root mean square error value equal to 2.6%. Based on the output of this model it was also possible to highlight the defective-maize areas within the images, confirming the significance of the proposed approach.
Due to the differences in terms of both price and quality, the availability of effective instrumentation to discriminate between Arabica and Robusta coffee is extremely important. To this aim, the use of multispectral imaging systems could provide reliable and accurate real-time monitoring at relatively low costs. However, in practice the implementation of multispectral imaging systems is not straightforward: the present work investigates this issue, starting from the outcome of variable selection performed using a hyperspectral system. Multispectral data were simulated considering four commercially available filters matching the selected spectral regions, and used to calculate multivariate classification models with Partial Least Squares-Discriminant Analysis (PLS-DA) and sparse PLS-DA. Proper strategies for the definition of the training set and the selection of the most effective combinations of spectral channels led to satisfactory classification performances (100% classification efficiency in prediction of the test set).
In the present paper, six categories of standard industrial grading tobacco provided by Hongta Group are taken as experimental samples, including three different tobacco locations-upper (B), middle(C) and lower(X) parts, with each part containing two kinds of tobacco colors-orange (O) and lemon yellow (L). Two methods including projection model method based on principal component and Fisher criterion (PPF) and support vector machine (SVM) method are used to analyze color and location features of tobacco based on visible-near infrared hyperspectral data. The results of projection model method indicate that in the projection and similarity analysis of tobacco color, location and six tobacco groups classified by color and location, two kinds of color can be fully differentiated, of which the similarity value is -1.000 8. Tobacco from upper and lower parts can also be fully differentiated with similarity value 0.405 3, but they both have intersections with tobacco from middle part. Six tobacco groups classified by color and location can be fully differentiated as well and their projection positions meet the actual external features of tobacco. The results of support vector machine method indicate that in the discriminant analysis of tobacco color, location and six tobacco groups classified by color and location, the average recognition rate of tobacco colors reaches 98%. The average recognition rate of tobacco location is 96%. The average recognition rate of six tobacco groups is 94%. Therefore, it’s feasible to analyze color and location features of tobacco using visible-near infrared hyperspectral data, which can provide reference for tobacco quality evaluation, computer-aided grading and tobacco intelligent acquisition, and also offers a new approach to the analysis of exterior features of other agricultural products.
Hyperspectral imaging can be an important tool for the assessment and documentation of the state of preservation of an object. Over time, documents that have experienced heavy usage will inevitably show evidence of handling, which can include staining. In this paper, the use of hyperspectral imaging is described for enhancing the assessment of the visual properties of stains. The use of imaging software (ENVI) is also described for quantitatively assessing the extent of staining in two different documents. Single 10nm bandpass images can be useful assessing darker stains with well defined boundaries. In one document (a treaty), the faint discolouration on one page made the extent of staining difficult to assess visually. A false colour density slice (450nm) provided a topographical image which was useful for enhancing the contrast between stained and unstained paper. In this type of image, the degree of discolouration could be correlated to optical density and the amount of staining on a page could then be related to the number of pixels for a given absorbance range. In a second document (a prayer book), the staining was more extensive and some of the stains were dark in appearance. This document also contained a lot of text that was written using a dark irongall ink, which limited the use of a density slice at a single bandpass. In this document, pixel unmixing was successfully used to quantitatively determine the extent of staining. The measurement tool provided with the Nuance™ Imaging System made it possible to quantitatively describe the size of the stain in terms of the number of pixels as well as its appearance in terms of average optical density.
Hyperspectral Imaging is an essential technique to deep explore surfaces in which more detail than the one provided by the single point spectroscopy is needed. Many devices for acquiring hyperspectral images have been manufactured and there is an increasing interest for improving the data analysis techniques applied to such complex datasets.Regardless the instrumentation, the acquisition of the images is being constantly improved by setting faster and more robust detectors, including new cooling systems or improving the light sources. Nevertheless, there are several issues that must be handled before starting the data analysis of any sample (e.g. background removal, compression of the images, spiked points, dead pixels, etc.). Therefore, the step of image pre-processing is almost always required.The aim of this paper is to show the application of some of the most common possibilities to solve the above mentioned issues before the image processing. This is done in a practical way, providing examples of their application, pros and cons as well as their implementation. For this purpose, several real examples (pharmaceutical tablets and food stuff) have been used throughout this manuscript.
A simple imaging system has been developed for acquiring multivariate images in order to characterise the heterogeneity of food materials. The objective of the present work is, first, to demonstrate the capability of this acquisition system to discriminate food products of different natures. Secondly, our goal is to apply Partial Least Squares regression on these multivariate images and to evaluate the interest of various strategies of classification. A data set containing 24 images (702 × 524) acquired at different wavelengths for four food products is analysed. After the establishment of the PLS2 models employed for predicting the indicator variables, four strategies of classification of observations are tested. The first classification is done by selecting the largest component of the indicator variables. The others are based on the measurement of distances to the barycentres of the qualitative groups. Distances calculated can be either Euclidian distances or Mahalonobis distances. Except the strategy based on the Euclidian distance on scores, the strategies are rather equivalent, with a slight advantage to the Euclidian distance on predicted indicators. Another possibility addressed by the use of linear discriminant analysis (LDA) on multivariate images is to represent the qualitative groups as artificial images. The largest confusion appears between both cereal products while others are well classified. Copyright © 2006 John Wiley & Sons, Ltd.
Techniques and Applications of Hyperspectral Image Analysis gives an introduction to the field of image analysis using hyperspectral techniques, and includes definitions and instrument descriptions. Other imaging topics that are covered are segmentation, regression and classification. The book discusses how high quality images of large data files can be structured and archived. Imaging techniques also demand accurate calibration, and are covered in sections about multivariate calibration techniques. The book explains the most important instruments for hyperspectral imaging in more technical detail. A number of applications from medical and chemical imaging are presented and there is an emphasis on data analysis including modeling, data visualization, model testing and statistical interpretation.
Banana fruit quality and maturity stages were studied at three different temperatures, viz., 20, 25, and 30 °C by using hyperspectral imaging technique in the visible and near infrared (400–1000 nm) regions. The quality parameters like moisture content, firmness and total soluble solids were determined and correlated with the spectral data. The spectral data were analyzed using the partial least square analysis. The optimal wavelengths were selected using predicted residual error sum of squares. The principal component analysis was also used to test the variability of the observed data. By using multiple linear regressions (MLR), models were established based on the optimal wave lengths to predict the quality attributes. The coefficient of determination was found to be 0.85, 0.87, and 0.91 for total soluble solids, moisture and firmness of the banana fruits, respectively. The change in TSS and firmness of banana fruits stored at different temperatures, viz., 20, 25, and 30 °C during the ripening process followed the polynomial relationships and the change in moisture content followed a linear relationship at different maturity stages.
This study was carried out to develop a hyperspectral imaging system in the near infrared (NIR) region (900–1700 nm) to assess the quality of cooked turkey hams of different ingredients and processing parameters. Hyperspectral images were acquired for ham slices originated from each quality grade and then their spectral data were extracted. Spectral data were analyzed using principal component analysis (PCA) to reduce the high dimensionality of the data and for selecting some important wavelengths. Out of 241 wavelengths, only eight wavelengths (980, 1061, 1141, 1174, 1215, 1325, 1436 and 1641 nm) were selected as the optimum wavelengths for the classification and characterization of turkey hams. The data analysis showed that it is possible to separate different quality turkey hams with few numbers of wavelengths on the basis of their chemical composition. The results revealed the potentiality of NIR hyperspectral imaging as an objective and non-destructive method for the authentication and classification of cooked turkey ham slices.
The effectiveness of Hyperspectral imaging (HSI) in the near infrared (NIR) range (1000–1700 nm) was evaluated to discriminate PET (polyethylene terephthalate) from PLA (poly(lactic acid)), two polymers commonly utilized as packaging for foodstuff, in order to improve their further recycling process. An internal calibration based on five reference materials was initially used to eliminate the variability existing among images, then Partial Least Squares-Discriminant Analysis (PLS-DA) was used to distinguish and classify the three classes, i.e., background, PET and PLA. Considering the high amount of data conveyed by the training image, the PLS-DA models were also calculated using as training set a reduced version of the original matrix, with the twofold aim to reduce the computational time and to deal with an equal number of spectra for each class, independently from the initial selected areas. A variable selection procedure by means of iPLS-DA was also applied on both the whole and the reduced matrix. The results obtained on the reduced matrix using only six variables provided a prediction efficiency higher than 98%. Moreover, the possibility to recognize PET and PLA polymers by HSI in the NIR range was further confirmed by using Multivariate Curve Resolution (MCR) as an alternative approach, which also allowed to evaluate the effect of thickness of the transparent plastic samples.
The dominant image processing tasks for hyperspectral data are compression and feature recognition. These tasks go hand-in-hand. Hyperspectral data contains a huge amount of information that need to be processed (and often very quickly) depending on the application. The discrete wavelet transform is the ideal tool for this type of data structure. There are applications that require such processing (especially feature recognition or identification) be done extremely fast and efficiently. Furthermore the higher number of dimensions implies a number of different ways to do these transforms. Much of the work in this area to the present time has been focused on JPEG2000 type compression of each component image involving fairly sophisticated coding techniques; relatively little attention has been paid to other configurations of wavelet transforms of such data, as well as rapid feature identification where compression may not be necessary at all. This paper describes other versions of the 3D wavelet transform that allow the resolution in both the spatial domain and spectral domain to be adjusted separately. Other issues associated with low complexity feature recognition with and without compression using versions of the 3D hyperspectral wavelet transforms will be discussed along with some illustrative calculations.
To detect various common defects on oranges, a hyperspectral imaging system has been built for acquiring reflectance images from orange samples in the spectral region between 400 and 1000nm. Oranges with insect damage, wind scarring, thrips scarring, scale infestation, canker spot, copper burn, phytotoxicity, heterochromatic stripe, and normal surface were studied. Hyperspectral images of samples were evaluated using principal component analysis (PCA) with the goal of selecting several wavelengths that could potentially be used in an in-line multispectral imaging system. The third principal component images using six wavelengths (630, 691, 769, 786, 810 and 875nm) in the visible spectral (VIS) and near-infrared (NIR) regions, or the second principal component images using two wavelengths (691 and 769nm) in VIS region gave better identification results under investigation. However, the stem-ends were easily confused with defective areas. In order to solve this problem, representative regions of interest (ROIs) reflectance spectra of samples with different types of skin conditions were visually analyzed. The researches revealed that a two-band ratio (R875/R691) image could be used to differentiate stem-ends from defects effectively. Finally, the detection algorithm of defects was developed based on PCA and band ratio coupled with a simple thresholding method. For the investigated independent test samples, accuracies of 91.5% and 93.7% with no false positives were achieved for both sets of selected wavelengths using proposed method, respectively. The disadvantage of this algorithm is that it could not discriminate between different types of defects.
A new graphically oriented local modeling procedure called interval partial least-squares (i PLS) is presented for use on spectral data. The i PLS method is compared to full-spectrum partial least-squares and the variable selection methods principal variables (PV), forward stepwise selection (FSS), and recursively weighted regression (RWR). The methods are tested on a near-infrared (NIR) spectral data set recorded on 60 beer samples correlated to original extract concentration. The error of the full-spectrum correlation model between NIR and original extract concentration was reduced by a factor of 4 with the use of i PLS (r=0.998, and root mean square error of prediction equal to 0.17% plato), and the graphic output contributed to the interpretation of the chemical system under observation. The other methods tested gave a comparable reduction in the prediction error but suffered from the interpretation advantage of the graphic interface. The intervals chosen by i PLS cover both the variables found by FSS and all possible combinations as well as the variables found by PV and RWR, and i PLS is still able to utilize the first-order advantage. Index Headings: Interval PLS; Variable selection; NIR, Principal variables; Forward stepwise selection; Recursively weighted regression; Beer; Extract.
This paper describes an approach for the colour-based classification of RGB images, taken with a common digital CCD camera on inhomogeneous food matrices. The aim was that of elaborating a feature selection/classification method independent of the specific food matrix that is analysed, in the sense that the variables that are the most relevant ones for the classification of the analysed samples are selected in a blind way, with no a priori assumptions on the basis of the nature of the considered food matrix. A one-dimensional signal describing the colour content of each acquired digital image, which we have called colourgram, is created as the contiguous sequence of the frequency distribution curves of the three red, green and blue colours values, of related parameters (also including hue, saturation and intensity) and of the scores values deriving from the PCA analysis of the unfolded 3D image array, together with the corresponding loadings values and eigenvalues. Once a sufficient number of digital images has been acquired, the corresponding colourgrams are then analysed by means of a feature selection/classification algorithm based on the wavelet transform, wavelet packet transform for efficient pattern recognition (WPTER). This approach was tested on a series of samples of “pesto”, a typical Italian vegetable pasta sauce, which presents high colour variability, mainly due to technological variables (raw materials, processes) and to the degradation of chlorophylls during storage. Good classification results (100% of correctly classified objects with very parsimonious models) have been obtained, also in comparison with the visual evaluation results of a panel test.
Hyperspectral imaging (HSI) combines spectroscopy and imaging resulting in three dimensional multivariate data structures (‘hypercubes’). Each pixel in a hypercube contains a spectrum representing its light absorbing and scattering properties. This spectrum can be used to estimate chemical composition and/or physical properties of the spatial region represented by that pixel. One of the advantages of HSI is the large volume of data available in each hypercube with which to create calibration and training sets. This is also known as the curse of dimensionality, due to the resultant high computational load of high dimensional data. It is desirable to decrease the computational burden implied in hyperspectral imaging; this is especially relevant in the development of real time applications. This paper gives an overview of some pertinent issues for the handling of HSI data. Computational considerations involved in acquiring and managing HSI data are discussed and an overview of the multivariate analysis methods available for reducing the considerable data load encountered in HSI data is presented.
A new approach for discrimination of objects on hyperspectral images, which combines state-of-art image processing methods and multivariate image analysis, is proposed. The basic idea of the approach is to build a joint principal component space for all objects' pixels, detect patterns, pixels from a particular object shared in this space, and use quantitative evaluation of the patterns as the objects' features. The approach was particularly developed for dealing with challenging cases, when objects from different classes have many similar pixels. It has been tested on several real cases and showed very promising results.
An increasing number of industrial applications requires visual inspection of products. Computer vision provides consolidated tools for reliable and fully automatic characterization and classification of the product quality at relatively low costs. One of such powerful tool is multivariate image analysis (MIA). In the MIA procedure as proposed in [1] is considered, that is well suited for texture analysis. To extend the performance of the MIA procedure in [1] to the analysis of wider spatial domains and to improve the algorithm from the computational point of view, a new formulation, named iMIA, has been recently proposed in [2]. The main contribution of the present paper is a modification of the iMIA algorithm that, by exploiting fast Fourier transform filtering, allows a considerable reduction of the computational time when spatial neighborhoods larger than few pixels are considered. Secondly, a different texture characterization with respect to [2] is proposed, to further extend the algorithm range of applicability. The characterization is based on histograms of textural features [3]. The algorithm is tested on two case studies in the field of texture analysis, namely, classification of rice quality, where the different characterization of texture allows a great improvement with respect to [2], and the characterization of nanofiber assemblies.
A poly(3,4-ethylendioxythiophene) modified electrode has been considered as a potentially useful amperometric sensor to use either alone or in the frame of a set of sensors bearing complementary information, i.e. within an electronic tongue. The sensor is proposed in blind analysis of red wines, for classification and calibration purposes. The data obtained from voltammetric measurements have been treated using partial least squares analysis. A calibration procedure has been performed to correlate results from analyses of wines, executed with traditional analytical methods, with the corresponding voltammetric responses. Moreover, classification models of the wine samples, based on quantitative parameters and qualitative information about origin and variety, have been built. The developed electrochemical sensor also allows the fast identification of samples exceeding threshold limits of meaningful parameters for quality control in the wine industry, such as SO2, colour intensity and total polyphenols. The application of the system within a sensor array (electronic tongue) to fast pre-screening routine control procedure is proposed.Highlights► An amperometric sensor recently developed by us is used on a wide number of red wines. ► Voltammograms are recorded on each sample. ► Calibration is performed between the compositional data and the voltammograms. ► Classification models for origin and variety are developed. ► Classification with respect to SO2, colour intensity and polyphenols is realised.
A novel algorithm based on coupling of the fast wavelet transform (FWT) with MLR and PLS regression techniques for the selection of optimal regression models between matrices of signals and response variables is presented: wavelet interface to linear modelling analysis (WILMA). The algorithm decomposes each signal into the FWT domain and then, by means of proper criteria, selects the wavelet coefficients that give the best regression models, as evaluated by the leave-one-out cross-validation criterion. The predictive ability of the regression model is then checked by means of external test sets. Moreover, the signals are reconstructed back in the original domain using only the selected wavelet coefficients, to allow for chemical interpretation of the results. The algorithm was tested on different literature data sets: two near-infrared data sets from Kalivas, on which the performances of many calibration algorithms have already been tested, and a data set consisting of lead and thallium mixtures measured by differential pulse anodic stripping voltammetry and giving seriously overlapped responses. Good results were obtained for all the studied data sets; in particular, for the data sets from Kalivas the WILMA models showed improved predictive capability. Copyright © 2003 John Wiley & Sons, Ltd.
Nowadays, image analysis is becoming more important because of its ability to perform fast and non-invasive low-cost analysis on products and processes. Image analysis is a wide denomination that encloses classical studies on gray scale or RGB images, analysis of images collected using few spectral channels (sometimes called multispectral images) or, most recently, data treatments to deal with hyperspectral images, where the spectral direction is exploited in its full extension. Pioneering data treatments in image analysis were applied to simple images mainly for defect detection, segmentation and classification by the Computer Science community. From the late 80s, the chemometric community joined this field introducing powerful tools for image analysis, which were already in use for the study of classical spectroscopic data sets and were appropriately modified to fit the particular characteristics of image structures. These chemometric approaches adapt to images of all kinds, from the simplest to the hyperspectral images, and have provided new insights on the spatial and spectroscopic information of this kind of data sets. New fields open by the introduction of chemometrics on image analysis are exploratory image analysis, multivariate statistical process control (monitoring), multivariate image regression or image resolution. This paper reviews the different techniques developed in image analysis and shows the evolution in the information provided by the different methodologies, which has been heavily pushed by the increasing complexity of the image measurements in the spatial and, particularly, in the spectral direction.
Hyperspectral imaging instruments produce large amounts of raw data. These raw data in A/D converter counts have a number of errors that can be corrected by calibration. The use of multiple Spectralon calibration standards is shown to correct for both spectral and spatial variations. Optimal results are achieved using a two-step calibration and correction process. A series of full field of view or external calibration standards is used to transform raw data counts to reflectance values. A grayscale series of internal standards embedded within each hyperspectral image is used to compensate for instrument instability. Second-order regression models based on these multiple standards provide maximum accuracy. The external standards allow for standardization within a hyperspectral image. The internal standards permit instrument standardization or calibration transfer between hyperspectral images. Copyright © 2006 John Wiley & Sons, Ltd.
Over the last two decades, near-infrared spectroscopy (NIRS) has established itself as a non-destructive analytical technique
in a variety of disciplines. However, recent technological advancements in hardware design and data mining techniques have
unleashed the potential of NIRS to become a tool of choice for routine analyses of agricultural products. The current paper
synthesizes the status of NIRS in the agri-food industry in terms of hardware and software development as well as the direction
in which the NIRS research is headed. An extensive review of literature reveals that the emphasis on hardware development
is focused on developing compact, robust, and portable spectrometers and hyperspectral imaging (HSI) systems. The software
development on the other hand is geared towards developing better preprocessing, analyses, and modeling techniques using chemometrics,
support vector machines, and artificial neural networks. The four main agri-food sectors identified to be the beneficiaries
of this research revolution are grain quality monitoring; post-harvest handling of fruits and vegetables; identification of
contaminants in animal produce and feed; and food safety and authenticity. Apart from discussing the aforementioned topics,
the paper also provides food scientists some working knowledge on parameters crucial to the performance of spectral and imaging
systems. It is expected that further development of NIRS will help agricultural and food scientists to enhance the quality
and safety of our food.
Hyperspectral imaging (HSI) is an emerging platform technology that integrates conventional imaging and spectroscopy to attain both spatial and spectral information from an object. Although HSI was originally developed for remote sensing, it has recently emerged as a powerful process analytical tool for non-destructive food analysis. This paper provides an introduction to hyperspectral imaging: HSI equipment, image acquisition and processing are described; current limitations and likely future applications are discussed. In addition, recent advances in the application of HSI to food safety and quality assessment are reviewed, such as contaminant detection, defect identification, constituent analysis and quality evaluation.
A PLS-bootstrap-VIP approach is proposed as a simple wavelength selection method, yet having the ability to identify relevant spectral intervals. This approach is particularly attractive for wavelength selection within hyperspectral images due to its simplicity and relatively low computational cost compared to more sophisticated interval search methods. The method was tested on four visible-NIR spectral imaging datasets taken from the polymer, oil and pulp and paper industries. The results were compared with those obtained using PLS regression coefficients as well as with two more sophisticated methods involving several metrics or search for wavelength intervals. It is shown that a small number of well defined relevant spectral intervals are identified with the proposed approach, providing easy spectral interpretation in agreement with more complex interval search methods. Before final use, fine adjustments to the VIP threshold may be tested to verify whether predictive power can be improved.
Merging spectroscopic imaging and chemometrics enhances the outcomes of instrumental technology and data analysis. Multivariate exploratory and resolution methods can be adapted to image analysis and provide global and local information about pure compounds in an imaged sample. Knowing in detail how the chemical compounds are distributed over the scanned surface gives valuable information about essential issues in the manufacture and the characterization of products, such as evenness of composition and, therefore, homogeneity of the sample. The power to detect and to locate impurities is also greatly enhanced because these unwanted compounds could show locally large concentrations (and signals), even though their abundance on the surface is very low. The capabilities of this combination are shown in an example of pharmaceutical product control, where analysis of the end product requires chemical characterization and quantitative information at global and local levels. The approach used and the kind of information obtained is general and can be applied to the analysis of images in other fields.
The emergence of chemical imaging (CI) has gifted spectroscopy an additional dimension. Chemical imaging systems complement chemical identification by acquiring spatially located spectra that enable visualization of chemical compound distributions. Such techniques are highly relevant to pharmaceutics in that the distribution of excipients and active pharmaceutical ingredient informs not only a product's behavior during manufacture but also its physical attributes (dissolution properties, stability, etc.). The rapid image acquisition made possible by the emergence of focal plane array detectors, combined with publication of the Food and Drug Administration guidelines for process analytical technology in 2001, has heightened interest in the pharmaceutical applications of CI, notably as a tool for enhancing drug quality and understanding process. Papers on the pharmaceutical applications of CI have been appearing in steadily increasing numbers since 2000. The aim of the present paper is to give an overview of infrared, near-infrared and Raman imaging in pharmaceutics. Sections 2 and 3 deal with the theory, device set-ups, mode of acquisition and processing techniques used to extract information of interest. Section 4 addresses the pharmaceutical applications.
Hyperspectral imaging in the visible and near-infrared (400–1000 nm) regions was tested for nondestructive determination of moisture content (MC), total soluble solids (TSS), and acidity (expressed as pH) in strawberry. The spectral data were analyzed using the partial least squares (PLS) analysis, a multivariate calibration technique. The correlation coefficients (r) with the whole spectral range (400–1000 nm) for predicting MC, TSS, and pH were 0.90, 0.80, and 0.87 with SEC of 6.085, 0.233, and 0.105 and SEP of 3.874, 0.184, and 0.129, respectively. Optimal wavelengths were selected using β-coefficients from PLS models. Multiple linear regression (MLR) models were established using only the optimal wavelengths to predict the quality attributes. The correlation coefficients (r) for predicting MC, TSS, and pH using MLR models were 0.87, 0.80, and 0.92 with SEC of 6.72, 0.220, and 0.084 and SEP of 5.786, 0.211, and 0.091, respectively. Moreover, for classifying strawberry based on ripeness stage, a texture analysis was conducted on the images based on grey-level co-occurrence matrix (GLCM). The higher classification accuracy of 89.61% was achieved using the GLCM parameters at horizontal direction at angle of 0°.
In the present paper, the possibility to use the information contained in RGB digital images to gain a fast and inexpensive quantification of colour-related properties of food is explored. To this aim, we present an approach which consists, as first step, in condensing the colour related information contained in RGB digital images of the analysed samples in one-dimensional signals, named colourgrams. These signals are then used as descriptor variables in multivariate calibration models. The feasibility of this approach has been tested using as a benchmark a series of samples of pesto sauce, whose RGB images have been used to predict both visual attributes defined by a panel test and the content of various pigments (chlorophylls a and b, pheophytins a and b, β-carotene and lutein). The possibility to predict correctly the values of some of the studied parameters suggests the feasibility of this approach for fast monitoring of the main aspect-related properties of a food matrix. The values of the squared correlation coefficient computed in prediction on a test set (R(Pred)(2)) for green and yellow hues were greater than 0.75, while R(Pred)(2) values greater than 0.85 were obtained for the prediction of total chlorophylls content and of chlorophylls/pheophytins ratio. The great flexibility of this blind analysis method for the quantitative evaluation of colour related features of matrices with an inhomogeneous aspect suggests that it is possible to implement automated, objective, and transferable systems for fast monitoring of raw materials, different stages of the manufacture and end products, not necessarily for the food industry only.
Hyperspectral chemical imaging (HCI) integrates imaging and spectroscopy resulting in three-dimensional data structures, hypercubes, with two spatial and one wavelength dimension. Each spatial image pixel in a hypercube contains a spectrum with >100 datapoints. While HCI facilitates enhanced monitoring of multi-component systems; time series HCI offers the possibility of a more comprehensive understanding of the dynamics of such systems and processes. This implies a need for modeling strategies that can cope with the large multivariate data structures generated in time series HCI experiments. The challenges posed by such data include dimensionality reduction, temporal morphological variation of samples and instrumental drift. This article presents potential solutions to these challenges, including multiway analysis, object tracking, multivariate curve resolution and non-linear regression. Several real world examples of time series HCI data are presented to illustrate the proposed solutions.
In the Italian oenological industry, the regular practice used to naturally increase the colour of red wines consists in blending them with a wine very rich in anthocyanins, namely Rossissimo. In the Asian market, on the other hand, anthocyanins extracted by black rice are frequently used as correctors for wine colour. This practice does not produce negative effects on health; however, in many countries, it is considered as a food adulteration. The present study is therefore aimed to discriminate wines containing anthocyanins originated from black rice and grapevine by using reliable spectroscopic techniques requiring minimum sample preparation. Two series of samples have been prepared from five original wines, that were added with different amounts of Rossissimo or of black rice anthocyanins solution, until the desired Colour Index was reached. The samples have been analysed by FT-NIR and (1)H NMR spectroscopies and the resulting spectra matrices were subjected to multivariate classification. Initially, PLS-DA was used as classification method, then also variable selection/classification methods were applied, i.e. iPLS-DA and WILMA-D. The classification with variable selection of NIR spectra permitted to classify the test set samples with an efficiency of about 70%. Probably these not excellent performances are due to the matrix effect, together with the lack of sensitivity of NIR with respect to minor compounds. On the contrary, very satisfactory results were obtained on NMR spectra in the aromatic region between 6.5 and 9.5 ppm. The classification method based on wavelet-based variables selection, permitted to reach an efficiency in validation greater than 95%. Finally, 2D correlation analysis was applied to FT-NIR and (1)H NMR matrices, in order to recognise the spectral zones bringing the same chemical information.
Until recently, applications of spectral imaging in heritage science mostly focused on qualitative examination of artworks. This is partly due to the complexity of artworks and partly due to the lack of appropriate standard materials. With the recent advance of NIR imaging spectrometers, the interval 1000-2500 nm became available for exploration, enabling us to extract quantitative chemical information from artworks. In this contribution, the development of 2D NIR quantitative chemical maps of heritage objects is discussed along with presentation of the first quantitative image. Further case studies include semiquantitative mapping of plasticiser distribution in a plastic object and identification of historic plastic materials. In the NIR imaging studies discussed, sets of 256 spatially registered images were collected at different wavelengths in the NIR region of 1000-2500 nm. The data was analyzed as a spectral cube, both as a stack of wavelength-resolved images and as a series of spectra, one per each sample pixel, using multivariate analysis. This approach is only possible using well-characterized reference sample collections, as quantitative imaging applications need to be developed, thus enabling spatial maps of damaged and degraded areas to be visualized to a level of chemical detail previously not possible. Such quantitative chemical mapping of vulnerable areas of heritage objects is invaluable, as it enables damage to historic objects to be quantitatively visualized.
Hyperspectral imaging under transmittance mode has shown potential for detecting internal defect, however, the technique still cannot meet the online speed requirement because of the need to acquire and analyze a large amount of image data. This study was carried out to select important wavebands for further development of an online inspection system to detect internal defect in pickling cucumbers and whole pickles. Hyperspectral transmittance/reflectance images were acquired from normal and defective cucumbers and whole pickles using a prototype hyperspectral reflectance (400-740nm)/transmittance (740-1000nm) imaging system. Up to four-waveband subsets were determined by a branch and bound algorithm combined with the k-nearest neighbor classifier. Different waveband binning operations were also compared to determine the bandwidth requirement for each waveband combination. The highest classification accuracies of 94.7 and 82.9% were achieved using the optimal four-waveband sets of 745, 805, 965, and 985nm at 20nm spectral resolution for cucumbers and of 745, 765, 885, and 965nm at 40nm spectral resolution for whole pickles, respectively. The selected waveband sets will be useful for online quality detection of pickling cucumbers and pickles.
Hyperspectral imaging techniques have widely demonstrated their usefulness in different areas of interest in pharmaceutical research during the last decade. In particular, middle infrared, near infrared, and Raman methods have gained special relevance. This rapid increase has been promoted by the capability of hyperspectral techniques to provide robust and reliable chemical and spatial information on the distribution of components in pharmaceutical solid dosage forms. Furthermore, the valuable combination of hyperspectral imaging devices with adequate data processing techniques offers the perfect landscape for developing new methods for scanning and analyzing surfaces. Nevertheless, the instrumentation and subsequent data analysis are not exempt from issues that must be thoughtfully considered. This paper describes and discusses the main advantages and drawbacks of the measurements and data analysis of hyperspectral imaging techniques in the development of solid dosage forms.
The use of near infrared (NIR) hyperspectral imaging and hyperspectral image analysis for distinguishing between hard, intermediate and soft maize kernels from inbred lines was evaluated. NIR hyperspectral images of two sets (12 and 24 kernels) of whole maize kernels were acquired using a Spectral Dimensions MatrixNIR camera with a spectral range of 960-1662 nm and a sisuChema SWIR (short wave infrared) hyperspectral pushbroom imaging system with a spectral range of 1000-2498 nm. Exploratory principal component analysis (PCA) was used on absorbance images to remove background, bad pixels and shading. On the cleaned images, PCA could be used effectively to find histological classes including glassy (hard) and floury (soft) endosperm. PCA illustrated a distinct difference between glassy and floury endosperm along principal component (PC) three on the MatrixNIR and PC two on the sisuChema with two distinguishable clusters. Subsequently partial least squares discriminant analysis (PLS-DA) was applied to build a classification model. The PLS-DA model from the MatrixNIR image (12 kernels) resulted in root mean square error of prediction (RMSEP) value of 0.18. This was repeated on the MatrixNIR image of the 24 kernels which resulted in RMSEP of 0.18. The sisuChema image yielded RMSEP value of 0.29. The reproducible results obtained with the different data sets indicate that the method proposed in this paper has a real potential for future classification uses.
This study investigated various chemical imaging methods for the forensic analysis of paints, tapes and adhesives, inks and firearm propellants (absorption and photoluminescence in the UV-vis-NIR regions). Results obtained using chemical imaging technology were compared with those obtained using traditional techniques. The results show that chemical imaging offers significant advantages in the forensic context, for example the ability to display visual and spectral results side by side and to reduce sample preparation, hence minimizing the risk of contamination. Chemical imaging produced a greater discriminating power than traditional techniques for most evidence types. Chemical imaging also eliminated different brands of ammunition based on the fluorescence characteristics of the propellant grains preserving the evidence for further analysis. It is expected that this technology will find broader forensic applications in the future.
A hyperspectral image in the near infrared contains thousands of position-referenced spectra. After imaging reference materials of known composition it is possible to build Partial Least Squares (PLS) regression models for predicting unknown compositions from new images or spectra. In this paper a comparison is made between spectra from a hyperspectral image and spectra from two spectrometers: a scanning grating instrument with rotating sample holders and an FT-NIR instrument utilizing a fiber-optic probe. The raw spectra and the quality of the PLS calibration models and predictions are compared. Two sample datasets consist of a set of 13 designed artificial mixtures of pure constituents and a selection of 13 sampled cheeses. The prediction error from the hyperspectral image spectra is between that of the two spectrometers. For a typical food sample, the average bias [and replicate standard deviation] was -0.6% [0.5%] for protein and -0.2% [1.3%] for fat. Comparable values for the best spectrometer were -0.2% bias for protein and -0.5% for fat. Some of the advantages of working with hyperspectral images are highlighted: the simultaneous exploration of representations of both spectral and spatial data, and the analysis of concentration profiles and concentration maps all contribute to better characterization of organic and biological materials.
Wavelets in Chemistry
- B Walczak
B. Walczak, Wavelets in Chemistry, first ed., Elsevier, Amsterdam, 2000.
- W Wang
- J Paliwal
W. Wang, J. Paliwal, Sens. Instrum. Food Qual. Saf. 1 (2007) 193-207.
- M Cocchi
- R Seeber
- A Ulrici
M. Cocchi, R. Seeber, A. Ulrici, J. Chemom. 17 (2003) 512-527.
- J Li
- X Rao
- Y Ying
J. Li, X. Rao, Y. Ying, Comput. Electron. Agric. 78 (2011) 38-48.
- S Chevallier
- D Bertrand
- A Kohler
- P Courcoux
S. Chevallier, D. Bertrand, A. Kohler, P. Courcoux, J. Chemom. 20 (2006) 221-229.
- J Burger
- P Geladi
J. Burger, P. Geladi, J. Chemom. 19 (2005) 355-363.
- G Elmasry
- A Iqbal
- D.-W Sun
- P Allen
- P Ward
G. ElMasry, A. Iqbal, D.-W. Sun, P. Allen, P. Ward, J. Food Eng. 103 (2011)
333-344.
- D Goltz
- M Attas
- G Young
- E Cloutis
- M Bedynski
D. Goltz, M. Attas, G. Young, E. Cloutis, M. Bedynski, J. Cult. Herit. 11 (2010)
19–26.
- P Williams
- P Geladi
- G Fox
- M Manley
P. Williams, P. Geladi, G. Fox, M. Manley, Anal. Chim. Acta 653 (2009) 121-130.
- J Burger
- A A Gowen
J. Burger, A.A. Gowen, Chemom. Intell. Lab. Syst. 108 (2011) 13-22.
- R Gosselin
- D Rodrigue
- C Duchesne
R. Gosselin, D. Rodrigue, C. Duchesne, Chemom. Intell. Lab. Syst. 100 (2010)
12-21.
- A Juan
- R Tauler
- R Dyson
- C Marcolli
- M Rault
- M Maeder
A. de Juan, R. Tauler, R. Dyson, C. Marcolli, M. Rault, M. Maeder, TrAC Trends
Analyt. Chem. 23 (2004) 70-79.
- J Burger
- P Geladi
J. Burger, P. Geladi, Analyst 131 (2006) 1152-1160.
- G Payne
- C Wallace
- B Reedy
- C Lennard
- R Schuler
- D Exline
- C Roux
G. Payne, C. Wallace, B. Reedy, C. Lennard, R. Schuler, D. Exline, C. Roux, Talanta
67 (2005) 334-344.