Remote Sensing Change Detection Based on Multi-scale Spatio-Temporal Perceptual Attention Network

Change detection in remote sensing images is an important application of remote sensing technology. Analyzing the temporal and spatial changes on Earth's surface can reveal the impacts of natural events or human activities. This paper proposes a change detection model with multi-scale feature extraction capabilities and spatiotemporal change relationship modeling. It employs Attentional Feature Fusion to design an AFF-ResNet Siamese Network for extracting and fusing features across multiple scales, thereby enabling the extraction of stronger discriminative features. Additionally, by combining Space-time attention, the model not only associates contextual information but also captures the differences between various images. Using cross-entropy loss as the loss function ensures faster convergence of the network. Finally, experiments conducted on the LEVIR-CD dataset validate the accuracy and effectiveness of the model.