Semantic segmentation of buildings in high-resolution remote sensing images based on DeepLabV3+ algorithm

Abstract

With the satellite remote sensing technology ushered in a leap of development, the resolution and clarity of satellite images have also been substantially improved, and high-resolution images depict the features more finely and provide more spectral information and texture contour information. The semantic segmentation of remote sensing image is one of the focuses of remote sensing technology research, which is very important for the development of remote sensing technology. To address the problems of imprecise target segmentation and low boundary segmentation accuracy in remote sensing image segmentation, a high-precision segmentation algorithm is proposed which based on DeepLabV3+. The algorithm optimizes the decoding region structure of the original network, adds the attention mechanism module, and improves the segmentation accuracy of remote sensing image.