Change Detection Network Based on Deformable Spatiotemporal Convolution

Land monitoring, disaster surveillance, and urban planning are currently key areas of focus. In recent years, with the rapid increase in the number of remote sensing satellites and significant advancements in imaging technologies, the volume of remote sensing image data has grown exponentially. Change detection, as a critical application of remote sensing imagery, holds substantial value in monitoring tasks. However, due to geometric transformations such as scale variations, rotations, and distortions in change targets, traditional convolutional methods often struggle to maintain robustness. To address this challenge, this paper proposes a change detection network based on deformable spatiotemporal convolution (DSTC-CD Net). The network adopts a dual-encoder and dual-decoder architecture, with its core innovation being the deformable spatiotemporal convolution module (DSTCM). This module generates offsets by integrating features from bi-temporal images. Compared with conventional convolution operations, it introduces temporal information and performs precise feature alignment through deformable convolution, effectively mitigating the impact of image deformation. Extensive experiments are conducted on three publicly available datasets-LEVIR-CD, CLCD, and SYSU-CD-to validate the effectiveness and superiority of the proposed method.