Research On Land Cover Monitoring Algorithm Based On Codec Structur

Monitoring land cover and utilization using high-resolution remote sensing images and extracting buildings and waters from urban and natural environments are important tasks.Accurate extraction can provide application value for environmental monitoring and land use.Therefore,extracting buildings and waters from high-resolution remote sensing images is one of the key tasks in land cover detection.Currently,most building and water area extraction algorithms use semantic segmentation methods based on deep learning,but there are still some shortcomings.On the one hand,targets such as buildings and water areas in remote sensing images have complex shapes and large differences in target sizes,which require full utilization of features at different scales.However,existing semantic segmentation methods still have shortcomings in utilizing multi-scale information.It is prone to problems such as large area misjudgment of buildings and blurred water edge.In order to solve this problem,this paper proposes a feature enhancement network based on a codec structure.In the decoding network,features of different scales are fused through a multiscale feature fusion module to obtain richer and more accurate feature representations,effectively improving problems such as large area misjudgment of buildings and blurring of water edges.On the other hand,due to the large amount of detail information in high-resolution remote sensing images,it is often difficult for local feature extractors to fully capture important features in the image,resulting in segmentation results such as errors and missing points.In order to solve the above problems,this paper proposes a local feature search network based on a codec structure.During the decoding process,integrating attention upsampling modules can enhance the search ability of local information,better capture the detailed features and texture information of the target,improve the recognition ability for small targets and occluded targets,and have a good application prospect.In comparative experiments,the algorithm proposed in this paper has achieved excellent segmentation results on both remote sensing datasets,with excellent performance in pixel accuracy,category average pixel accuracy,and average intersection and merge ratio.