Unsupervised change detection using Spatial Transformer Networks

Change detection in images or video sequences is used to detect anomalies or important variations in the scene. Although much research has focused on change detection in surveillance videos, there is little work related to change detection on aerial or satellite imagery. This paper presents a deep learning approach for change detection in satellite images. The proposed approach utilizes Spatial Transformer Networks (STNs) which learn to perform a coordinate transformation to localize potential change regions. With unsupervised training the STN system examines differences between images for significant change activity. Our initial results illustrate the merit of this method.