MSCFNet: A Multi-scale Spatial and Channel Fusion Network for Geological Environment Remote Sensing Interpreting

With the rapid development of Geological Environment Remote Sensing (GERS) technology, accurately interpreting geological elements has become a critical task in the fields of geology and environmental science. To address the issue of low model interpretation accuracy caused by intra-class variation, inter-class similarity, and complex distribution in GERS, a new Multi-Scale Spatial and Channel Fusion Network MSCFNet, which consists of the Fine-grained Local feature Fusion (FLF) module, Multi-resolution Geological Context-Aware (MGCA) module, and Global Feature Aggregation (GFA) module, are proposed. A series of experiments on the GERS dataset of Northwest China have demonstrated the significant advantages of our approach. Compared with the mainstream semantic segmentation model, it has improved mPA by 3.1% and mIoU by 3.32%. Additionally, ablation experiments are performed to verify the performance enhancement of each module.