A Light-Weight Neural Network Using Multiscale Hybrid Attention for Building Change Detection

The study of high-precision building change detection is essential for the sustainable development land resources. However, remote sensing imaging illumination variation and alignment errors have a large impact on accuracy detection. A novel lightweight Siamese neural network model proposed error problem caused by non-real changes in high-resolution images. feature extraction module acquires local contextual information at different scales, allowing it to fully learn global features. hybrid attention consisting channel spatial can make full use rich spatiotemporal semantic around achieve accurate changing buildings. For problems span which easily lead rough edge details missed small-scale buildings, multi-scale concept introduced divide extracted maps into multiple sub-regions introduce separately, finally, output features scales are weighted fused enhance detail capability. was experimented WHU-CD LEVIR-CD public data sets achieved F1 scores 87.8% 88.1%, respectively, higher than six comparison models, only cost 9.15 G MACs 3.20 M parameters. results show that our while significantly reducing number