Efficient Lightweight Remote Sensing Change Detection Using GPU-Based Computation

High-resolution remote sensing (RS) change detection (CD) plays a vital role in monitoring surface changes, with applications in disaster management, urban expansion, and military operations. However, challenges such as feature convergence and false changes induced by seasonal variations and lighting differences hinder accurate detection. To address these issues, we propose a lightweight differential convolution operation method, a dual-temporal change detection approach that incorporates multi-scale convolutions and differential operations. Specifically, we design a convolutional module consisting of multiple convolution blocks at different scales, interconnected by residual connections, to enhance the capture of temporal features across different scales. Additionally, we introduce differential operations that help eliminate false changes, resulting in enhanced differential feature maps. This method strengthens the model's ability to capture critical information across different time periods and effectively mitigates false change detection. This lightweight approach reduces GPU computational pressure while maintaining model effectiveness, enabling GPU-based computation in resource-constrained environments.