A SIAMESE DIFFERENTIAL NETWORK WITH GLOBAL CONTEXT ENHANCEMENT FOR BITEMPORAL REMOTE SENSING IMAGES CHANGE DETECTION

Change detection (CD) of remote sensing images aims to identify change areas from bitemporal image pairs. In recent years, with the help of a large amount of remote sensing data and the explosive development of artificial intelligence technologies, deep learning methods represented by Convolutional Neural Networks have brought remarkable success in the field of change detection. However, existing methods have difficulty in detecting the detailed change information effectively. In this paper, we propose a siamese differential network with global context enhancement based on the common fully convolution network architecture, namely SDGC-FCN. In the encoding stage, bitemporal images are separately fed into a weight-shared encoder network by differencing feature maps at different scales. In the decoding stage, the difference maps at different levels are concatenated and further processed by a spatial-temporal context enhancement module to highlight the variation regions and obtain significant difference feature maps. Quantitative and qualitative results conducted on LEVIR-CD and WHU-CD datasets demonstrate the effectiveness of our proposed method.