Remote Sensing Image Change Detection Based on Fully Convolutional Network With Pyramid Attention

Change detection technology based on remote sensing image can monitor the changes of ecological environment, which is of great significance for the study of the interaction between human and natural environment. However, it's difficult to automatically mine useful change information in traditional methods with the explosive growth of remote sensing data. With the development of deep learning methods, change detection of remote sensing images based on fully convolutional neural network has become one of the research hotspots to extract change information automatically. Recently, two advanced deep learning models, FC-EF and FC-Siam-diff, have been proposed for change detection of bi-temporal remote sensing images. To dig deeper multiscale and multilevel features thoroughly and improve detection accuracy, pyramid attention layer is added and an improved fully convolutional network FC-Siam-diff-PA is proposed in this paper. By introducing pyramid attention layer, the multiscale change information is further extracted from the difference feature map processed by the encoder structure of original network. Experiments were performed in the region of Xishuangbanna, Yunnan Province, China. Experimental results show that the proposed method performs better than the FC-EF network model and FC-Siam-diff model in change information extraction.