The Image and Signal Processing (ISP) group at the Universitat de València has harmonized a big database of labeled multi- and hyperspectral images for testing classification algorithms. We think that, like in other related fields of science, data sharing and reproducibility are the only ways for fostering true advance in remote sensing data processing. So far we have harmonized 43 image datasets, both multi- and hyperspectral. We want to expand this database as much as possible in order to objectively evaluate algorithms and submitted papers. We provide training pairs (spectra and their labels) and test spectra. Researchers are able to train their algorithms off-line, and then evaluate their accuracy over an independent, fixed, spectra test set per image. The system returns accuracy and robustness measures of your algorithm in that test set, as well as a ranked list of the best methods. The datasets and the automatic testing system will be available as soon as no data copyright conflict is identified.

ALTB is a set of tools implementing state-of-the-art active learning algorithms for remote sensing applications.

A simple Toolbox for Anomaly Change Detection (ACD) with Gaussianity assumptions and Elliptically Contoured (EC) distributions.

A set of train-test simple educational functions for data classification: LDA, QDA, Mahalanobis-distance classifier, decision trees, random forests, SVM, Boosting, Bagging, Gaussian process classifiers, etc.

Last version of the toolbox is in GitHub: https://github.com/IPL-UV/simpleClass.

A graph kernel for spatio-spectral remote sensing image classification with support vector machines (SVM). The method considers higher order relations in the neighborhood (beyond pairwise spatial relations) to iteratively compute a kernel matrix for SVM learning. The proposed kernel is easy to compute and constitutes a powerful alternative to existing approaches.

A graph-based method for semi-supervised learning: essentially an affinity matrix is computed, the graph Laplacian is normalized, and a spreading function is iterated until convergence. This algorithm can be understood intuitively in terms of spreading activation networks from experimental psychology, and explained as random walks on graphs. We successfully apply it to hyperspectral image classification. It incorporates contextual information through a full family of composite kernels. Noting that the graph method relies on inverting a huge kernel matrix formed by both labeled and unlabeled samples, we originally introduce the Nyström method in the formulation to speed up the classification process.

A semi-supervised SVM for the classification of remote sensing images. The method exploits the wealth of unlabeled samples for regularizing the training kernel representation locally by means of cluster kernels. The method learns a suitable kernel directly from the image, and thus avoids assuming a priori signal relations by using a predefined kernel structure. The method scales almost linearly with the number of unlabeled samples and provides out-of-sample predictions.

A (large margin) SVM algorithm that learns convolutional filters, with applications to time series analysis and remote sensing image classification.

This code+data snippet implements the automatic change detection algorithm presented in the SPIE 2010 and IGARSS 2011 papers. It consists in three steps: initialization (histogram-based), automatic parameter tuning and change map generation, with classical kmeans, gaussian kernel kmeans and by clustering the difference image in the feature spaces. Loops allows to perform different experiments and the user can choose different parameters (number of experiments, kernel function and parameters to search, number of pseudo training samples,...). Moreover, one can easily adapt the code for comparisons to other algorithms and images, as well as adapting the code for personal developments.