Fully Convolutional Network-Based Change Detection in Remote Sensing Images

The technological progress in the computer vision field along with the new generations of space borne remote sensing sensors has revolutionized the remote sensing science, especially in change detection task. In this investigation, a fully convolutional neural network-based remote sensing images change detection framework is introduced. The proposed approach is an unsupervised one, which exploits the features extracted from different layers of a deep neural network without pooling layers. The evaluation of this approach is performed using pair of images from the ONERA change detection dataset, by considering established performance criteria. These experiments show very promising quantitative and visual results provided by the proposed approach, and illustrate its efficiency for change detection in urban and agricultural areas.