Sentinel-2 Change Detection Based on Deep Features

In this manuscript, we address the problem of change detection for Sentinel-2 data. The proposed method is based on deep features representation. First, multilevel convolutional neural network (CNN) features are extracted from input images acquired at different times. Then, euclidean distance is applied to generate dissimilarity map that indicate change probabilities of each pixel. Finally, bounding boxes corresponding the change areas can be obtained with clustering and an optimizing connected component labeling algorithm. Experiments on a manually annotated dataset demonstrate the feasibility and effectiveness of the proposed method.