Water Body Segmentation Model And Comprehensive Evaluation Of Rural Water System Connectivity Based On Deep Learning

Water system identification has been an important research area,which is significant to include water resources allocation,water ecology,urban flood prevention and mitigation,and water beauty countryside construction.The goal of water body identification is to accurately delineate the extent of water bodies such as lakes,rivers and wetlands based on remote sensing images.In recent years,remote sensing image feature recognition based on deep learning semantic segmentation techniques has been far ahead of traditional machine learning recognition methods in recent years of development,but in general,there is still room for improvement,especially in dealing with the challenges posed by cloud or feature occlusion and background noise interference,and in developing techniques that can effectively identify small or fragmented water bodies.This paper therefore focuses on further research in this area to advance the latest technological developments in water body identification and to achieve more accurate and efficient monitoring of dynamic water systems.In this paper,14 natural villages in Hezheng County,a representative county in the second batch of pilot counties for the construction of water beauty villages,are selected as the study area.By introducing several design modules to improve the traditional segmentation model,the problem of difficult to accurately segment small water systems in villages is effectively solved and water body information can be accurately extracted.Combined with Arc GIS,spatial superposition analysis,spatial autocorrelation analysis,hierarchical analysis and other tools and methods to achieve the extraction of attribute information of water bodies,spatial distribution of water bodies,spatial and temporal evolution analysis of river channels and multi-index water body connectivity evaluation.Initially,the following research results were obtained:(1)Through relevant software tools,the Google satellite image data of 14 natural villages in the pilot area of Hezheng County in 2013 and 2022 were extracted,semantically annotated,cropped into image files and label files of the same size to obtain a total of 2882 images of water body identification dataset,and the data were randomly disrupted and re-cut into training set,validation set and test set.(2)For the current semantic segmentation model constructed by encoder and decoder,this study designs a scene perception module with multiple parallel structures,which increases the perceptual field and can capture multi-scale features,and reduces the number of parameters by separable convolution to accelerate the network training;introduces a dual attention module at the top layer of the network to retain and enhance the respective information from both location and channel,ignores The water body guidance module is designed to compensate for the detail information lost in the downsampling process and guide the fusion of the high-level feature information with the information in this layer,thus enhancing the extraction of water body information while ignoring the background influence to achieve the purpose of refined segmentation;in addition,to verify the feasibility of the module,ablation experiments are designed to observe the model performance after adding each module,so that the results of this network model In addition,to verify the feasibility of the module,the ablation experiment was designed to observe the performance of the model after adding each module,so that the results of the network model are scientific and rigorous.(3)Use Arc GIS to vectorize the segmented data and extract the center of the river to get the basic attributes of the water body.The basic data,spatial pattern indicators and spatial overlay analysis of the two phases show that the overall water body changes in the study area in the two phases are mainly influenced by the widening of the channel of the main stream Da Nancha River and the tributary Xiao Xia River,and human activities have less influence on the overall environment;the center of gravity migration model and the standardized ellipse show that the overall water body area change direction distribution is consistent with the standardized ellipse,which is distributed around the Da Nancha River and Xiao Xia River,and the The results of the center of gravity analysis showed that the center of gravity of area change moved 483.60 m to the northeast;using the global Moran index and outliers in the spatial autocorrelation analysis and cluster analysis,it was found that the amount of river change in Lang Tuquan village among 14 villages was smaller than that in the two neighboring villages,showing a low-high local spatial distribution pattern,but since the results of the global Moran index showed a random distribution,in order to test the In order to test the results of local analysis,the cold hot spot analysis method was used to verify that Lang Tuquan village is a hot spot area in spatial distribution,which indicates that the overall amount of river channel changes in it and the two neighboring villages is larger compared with other places,verifying the local non-random distribution and forming a spatial agglomeration mainly in Wolf Tuquan;finally,12 index factors were designed from three aspects of village water system,namely,life,ecology and safety,through the hierarchical analysis method.The comprehensive evaluation of water system connectivity in Hezheng County is in the transition stage from medium to good.The above research results provide new methods and new ideas for water system extraction,water system evolution and water system evaluation,and also lay a solid theoretical foundation for the future development of water beauty countryside construction.