Application of CSGE-PSPnet Remote Sensing Image Semantic Segmentation Technology in Transmission and Transformation Engineering Design

With the breakthrough development of remote sensing technology and deep learning, remote sensing technology also has a broad application prospect in the field of transmission and transformation engineering design. Semantic segmentation, as a key link in remote sensing image processing, is the basis for improving the intelligence of transmission and substation engineering, for siting, construction, maintenance, and monitoring of transmission and substation engineering by accurately identifying important facilities, sensitive locations and environmental changes. Therefore, this paper proposes a remote sensing image semantic segmentation model(CSGE-PSPnet), which is based on PSPNet(Pyramid Scene Parsing Network) and SGE(Spatial Group-Wise Enhance Attention). CSGE-PSPnet could enhance local significance features while extracting global features, realizes the refined extraction of key information, and designs CSGE Block(Spatial-Channel Group-Wise Enhance Attention) to extract channel and spatial local information at the same time to enhance local significance features. Multi-level features are fused through dense connection structures to retain feature information of small-scale targets. Subsequently, the deep global features are mined by PSPNet on the feature map after fusion, and the contextual information of different regions is fused to obtain the segmentation results. In addition, one auxiliary loss layer and one major loss layer further improve the model's generalization ability. The experimental results show that CSGE-PSPnet is superior to other methods and can accurately identify small targets, and CSGE Block significantly enhances the feature representation.