Research On Land Cover Segmentation Algorithm Based On Dual-branch Parallel Structur

After decades of evolution,the land cover classification method in high-definition satellite maps was gradually improved.Recently,high-definition remote sensing maps were applied to land cover classification.At present,land cover classification methods for high-resolution remote sensing images still need to be improved.On the one hand,with the development of edge computing and the higher resolution of remote sensing image,the efficiency of real-time semantic segmentation is limited by the huge amount of computation and parameter.Therefore,this paper proposes a semantic guided bottleneck network based on a dual-branch parallel structure.It improves the reasoning efficiency of the network through the redesign of the basic unit and the optimization of the overall structure of the network,so that it can extract spatial details and contextual semantic information efficiently.And restore high-resolution pixel-level features with a lightweight attention mechanism.On the other hand,high-resolution remote sensing images produce more complex information,and different image acquisition sources also cause data interference,which leads to the semantic segmentation model of land cover semantic segmentation can not effectively focus on the key information.Therefore,this paper proposes a local and global feature coupling network based on a dual-branch parallel structure.It uses convolution operators to extract local features and self-attention mechanisms to capture global representations to enhance perceptual learning.Local features and global representations with semantic divergence are fused through a coupling module.Finally,a decoder is designed to maximize the preservation of local features and global representations,and to better recover high-definition feature maps.Finally,it is proved by experiments that the two algorithms proposed in this paper achieve satisfactory results in land cover recognition.