DAFT: Differential Feature Extraction Network Based on Adaptive Frequency Transformer for Remote Sensing Change Detection

Remote sensing change detection is an important research direction in the field of remote sensing. It mainly used to focus on changing information ground over a period time, and identify interested targets from it. The rapid changes due social development undoubtedly increase importance detection. Currently, methods still have some shortcomings dealing with complex targets, environmental noise, other aspects. Therefore, we propose differential feature extraction network based adaptive frequency transformer for (DAFT). Adaptive (AFFormer) capable separating environments perspective capturing long-range dependencies between through self-attention. DAFT, use AFFormer as backbone extract bitemporal images, enhancing our while obtaining richer more detailed information. To knowledge, this first time that has been applied CD. address issues missing location insufficient local correlation, DAFT proposes features enhancement module reconstruction stage targets. In addition, uses DO-Conv enhance pixel correlation calculation convolutional operations, allowing By outputting results at different scales during stage, computes multiple losses are summed up guide training process better performance. experimental prove achieves high versus mainstream networks. On LEVIR-CD F1 91.814 IoU 84.866; WHU-CD 92.085 85.330; GZ-CD 86.065 74.512.