Change Detection of Open Pit Mining Areas in Remote Sensing Images Based on Coupled Local-Global Features

In this paper, a coupled local-global feature-based change detection method Swin-LGCNet is proposed for remote sensing image open pit mining areas The model interactively fuses local and global features extracted by ResNet and Swin Transformer through the local-global feature coupling module, which effectively improves the feature expression capability. The Mish activation function is used instead of ReLU, which reduces the information loss and enhances the model's ability to capture complex changing features. Meanwhile, the differential operation is introduced to further enhance the difference between the dual-time phase features and significantly reduce the leakage detection rate. The experimental results show that Swin-LGCNet outperforms the traditional network in both global information extraction and local detail performance, and the detection accuracy is significantly improved compared with the comparison networks, such as PSPNet, Deeplabv3+, UNet and UNet++, especially in boundary identification and noise suppression. The method makes full use of the temporal sequence of remote sensing images, and provides efficient technical support for dynamic change monitoring and ecological protection in open pit mining areas.