Fig 2 - uploaded by Ribana Roscher
Content may be subject to copyright.
Results obtained for hyperspectral plant dataset. (a) Image data, (e) Image data, detail, (b) and (f) Clustering result with k-means on original representation, (c) and (g) Clustering result with k-means on sparse representation coefficients, (d) and (h) Spectral clustering result on a sparse representation-based archetypal graph. Colors are arbitrary and indicate assignment to clusters. Therefore, they are not related between the images. 

Results obtained for hyperspectral plant dataset. (a) Image data, (e) Image data, detail, (b) and (f) Clustering result with k-means on original representation, (c) and (g) Clustering result with k-means on sparse representation coefficients, (d) and (h) Spectral clustering result on a sparse representation-based archetypal graph. Colors are arbitrary and indicate assignment to clusters. Therefore, they are not related between the images. 

Source publication
Conference Paper
Full-text available
We propose sparse representation-based archetypal graphs as input to spectral clustering for anomaly and change detection. The graph consists of vertices defined by data samples and edges which weights are determines by sparse representation. Besides relationships between all data samples, the graph also encodes the relationship to extremal points,...

Similar publications

Article
Full-text available
Recent advancements in learning from graph-structured data have shown promising results on the graph classification task. However, due to their high time complexities, making them scalable on large graphs, with millions of nodes and edges, remains a challenge. In this paper, we propose NetKI, an algorithm to extract sparse representation from a giv...
Preprint
Full-text available
Recent work in graph neural networks (GNNs) has lead to improvements in molecular activity and property prediction tasks. However, GNNs lack interpretability as they fail to capture the relative importance of various molecular substructures due to the absence of efficient intermediate pooling steps for sparse graphs. To address this issue, we propo...

Citations

... Numerous techniques [19]- [22], [27]- [28] have been developed to evaluate the similarities among data points. Among these, sparse representation stands out, especially noted for its superior performance in managing highdimensional or sparse datasets. ...
... Thus, to effectively utilize both local and global information, we suggest a method based on sparse representation [27], [28] to determine the weights of each data point. Drawing from [22], we define the objective function of our method as follows: ...
Article
Full-text available
The approach to calculating density significantly impacts the clustering efficacy of the Density Peak Clustering (DPC) method, with various density calculation methods tailored for different datasets. To address this, this study introduces a Self-Representation Weighted Density Peak Clustering (SR-DPC) method. Unlike traditional DPC, SR-DPC not only utilizes the local data point information but also amplifies the impact of different data points on the data center by implementing a weighting strategy, enhancing the precision with which data centers are identified. Moreover, SR-DPC adaptively reflects the influence of diverse data points on the data centers via feature representation and employs a weighted Gaussian kernel distance instead of the Euclidean distance to boost its clustering capabilities. Experimental evaluations on both synthetic and real datasets demonstrate the effectiveness and practicality of the SR-DPC method.
... So far, many methods [16][17][18][19][24][25] are proposed to measure the similarities of all the data points. In these methods, sparse representation is a typical resolution and has been confirmed its outperformed performance on the high-dimensional and/or highly sparse datasets. ...
... Therefore, in order to make full use of the local and the global information, we propose a sparse representation-based method [24,25] to calculate the weight of each data point, referring to [19], the objective function of the proposed method is defined as follows: ...
Preprint
Full-text available
The calculation method of density has an important influence on the clustering performance of density peak clustering method (DPC), the different density calculation methods are applied for the different datasets. To solve this problem, a self-representation weighted density peak clustering method (SR-DPC) is proposed in this study. Different from DPC, SR-DPC not only considers the local information of data points, but also enhances the influence of different data points on the data center by introducing the idea of weighting, so as to improve the accuracy of finding the data center.Furthermore, SR-DPC can adaptively reflect the influence of different data points on the data center points through the feature representation among data points, and the weighted Gaussian kernel distance is adopted to replace the Euclidean distance, so as to improve its clustering performance. Experimental results based on synthetic datasets and real datasets show that SR-DPC method is effective and practical.
... The same methodology, improved by Yang et al. in [54] by applying deep canonical correlation analysis, could not perform better than 0.947 (standard deviation 0.02) in OA and 0.71 (standard deviation 0.1) in KC. In [55], Roscher et al. achieved a KC of 0.80 on the same dataset but by using two Landsat 5 TM images, so performing homogeneous CD. Overall, our unsupervised method consistently outperforms the aforementioned methods, both in terms of OA and KC. ...
Preprint
Full-text available
Change detection in heterogeneous multitemporal satellite images is an emerging and challenging topic in remote sensing. In particular, one of the main challenges is to tackle the problem in an unsupervised manner. In this paper we propose an unsupervised framework for bitemporal heterogeneous change detection based on the comparison of affinity matrices and image regression. First, our method quantifies the similarity of affinity matrices computed from co-located image patches in the two images. This is done to automatically identify pixels that are likely to be unchanged. With the identified pixels as pseudo-training data, we learn a transformation to map the first image to the domain of the other image, and vice versa. Four regression methods are selected to carry out the transformation: Gaussian process regression, support vector regression, random forest regression, and a recently proposed kernel regression method called homogeneous pixel transformation. To evaluate the potentials and limitations of our framework, and also the benefits and disadvantages of each regression method, we perform experiments on two real data sets. The results indicate that the comparison of the affinity matrices can already be considered a change detection method by itself. However, image regression is shown to improve the results obtained by the previous step alone and produces accurate change detection maps despite of the heterogeneity of the multitemporal input data. Notably, the random forest regression approach excels by achieving similar accuracy as the other methods, but with a significantly lower computational cost and with fast and robust tuning of hyperparameters.
... The same methodology, improved by Yang et al. in [54] by applying deep canonical correlation analysis, could not perform better than 0.947 (standard deviation 0.02) in OA and 0.71 (standard deviation 0.1) in KC. In [55], Roscher et al. achieved a KC of 0.80 on the same dataset but by using two Landsat 5 TM images, so performing homogeneous CD. Overall, our unsupervised method consistently outperforms the aforementioned methods, both in terms of OA and KC. ...
Article
Full-text available
Change detection (CD) in heterogeneous multitemporal satellite images is an emerging and challenging topic in remote sensing. In particular, one of the main challenges is to tackle the problem in an unsupervised manner. In this paper, we propose an unsupervised framework for bitemporal heterogeneous CD based on the comparison of affinity matrices and image regression. First, our method quantifies the similarity of affinity matrices computed from colocated image patches in the two images. This is done to automatically identify pixels that are likely to be unchanged. With the identified pixels as pseudotraining data, we learn a transformation to map the first image to the domain of the other image and vice versa. Four regression methods are selected to carry out the transformation: Gaussian process regression, support vector regression, random forest regression (RFR), and a recently proposed kernel regression method called homogeneous pixel transformation. To evaluate the potentials and limitations of our framework and also the benefits and disadvantages of each regression method, we perform experiments on two real data sets. The results indicate that the comparison of the affinity matrices can already be considered a CD method by itself. However, image regression is shown to improve the results obtained by the previous step alone and produces accurate CD maps despite of the heterogeneity of the multitemporal input data. Notably, the RFR approach excels by achieving similar accuracy as the other methods, but with a significantly lower computational cost and with fast and robust tuning of hyperparameters.