An Unsupervised Siamese Superpixel-Based Network for Change Detection in Heterogeneous Remote Sensing Images

In this paper, we consider the problem of change detection in heterogeneous remote sensing images. Existing deep learning-based methods for change detection often utilize square convolution receptive fields, which do not sufficiently exploit the contextual information in heterogeneous images. Square receptive fields reduce the robustness to change detection scenarios with complex contextual structures, increase the number of false alarms, and degrade the performance of change detection. To address the aforementioned issue, we propose an unsupervised Siamese superpixel-based network (US ² N) for change detection in heterogeneous remote sensing images. Our newly proposed method innovatively combines superpixels with the square receptive fields to generate the boundary adherence receptive fields and better capture the contextual information than existing methods only with the regular square receptive fields. Experiments based on two real data sets demonstrate that the proposed method achieves higher accuracy than other commonly used change detection methods in heterogeneous remote sensing images.