A Novel Joint Change Detection Approach Based on Weight-Clustering Sparse Autoencoders

With the rapid development of earth observation technology, the number of available remote sensing data has soared dramatically. It becomes a significant problem that how to use remote sensing images and how to improve the accuracy of change detection effectively. In this paper, a novel approach for change detection using weight-clustering sparse autoencoders (WCSAE) combined object-oriented classification with difference images (DIs) is proposed. First, bi-phase images are segmented as patches through the density-based spatial clustering of applications with noise algorithm. Afterward, the average and variance of superpixels are stacked as the input of WCSAE. To reduce the redundant information of extracted features, similar weights in the hidden layer of WCSAE are clustered layer-wise under termination conditions by using the hierarchical agglomerative clustering algorithm. For the improvement of the classification accuracy, L1/2 regularization is introduced in the objective function to extract more sparse features and avoid over-fitting. Next, the post-classification change detection map is obtained by means of comparing with classification results of two phase images. Then, by using the change vector analysis technology, the difference map is yielded and also classified under WCSAE to acquire the DI classification map. Finally, the joint probability judgment is implemented on the joint scopes to determine changed and unchanged areas. The effectiveness and superiority of the proposed method are verified in accordance with the experimental results on standard datasets and actual remote sensing images.