Research On The Extraction And Spatiotemporal Evolution Of Rural Settlements In Liangzhou District,Wuwei City Based On Deep Learning

As the core of the interaction between rural people and land and the carrier of regional culture,the planning and development of rural settlements are crucial for the sustainable development of rural areas.However,with the accelerated progress of new urbanization and industrialization,rural development in China generally faces problems such as low land use efficiency and scattered layout.At the same time,the increasing number of rural population flowing to cities has led to an imbalance in the rural land structure and an intensification of rural hollowing out.In the arid areas of northwest China,the problems in rural areas are more prominent due to backward economic development and fragile ecological environment.Therefore,in order to effectively utilize rural land and resources and optimize the spatial structure of rural residential land,it is necessary to accurately grasp the information of rural residential land and the spatiotemporal distribution characteristics and driving factors of rural residential land,in order to provide scientific and effective support for the new rural construction and rural reform under the rural revitalization strategy.Traditional methods for extracting rural residential areas rely on manually designed features,resulting in low extraction accuracy and limited generalization ability.With the emergence of high-resolution remote sensing data and deep learning technologies,new data foundations and technical support have been provided for the rapid and accurate extraction of rural residential areas.This article takes Liangzhou District,located in the arid and semi-arid region of northwest China,with a fragile ecological environment and high population density,as the research object.Using multi-source high-resolution remote sensing image data,based on the classic convolutional neural network U-Net,a self attention module is introduced to establish a remote sensing image rural residential area extraction model.On the basis of the rural residential area information extracted by the model,the spatial and temporal change characteristics of rural residential areas in Liangzhou District from 2010 to 2020 are explored by integrating GIS spatial analysis and landscape pattern index,and the influencing factors of the driving factors of rural settlements spatial differentiation are quantitatively analyzed by using the geographical detector model.The specific research content and conclusions are as follows:(1)Construct rural settlements extraction dataset.This article is based on Google high-resolution image data from Liangzhou District in 2010,2015,and 2020 for manual labeling,with a total labeled image range of about 1200 km~2.After data processing and data enhancement,a total of 30939 sheets and 256×256 samples constitute a multi temporal semantic segmentation dataset for rural settlements,which can be used for training and verifying semantic segmentation models for rural residential areas in northwest China.(2)Build based on improvement U-Net network based rural settlements extraction model.By comparing with Seg Net and U-Net models on WHU data set and self-built data set,the results show that the improved model based on U-Net network proposed in this paper has excellent performance,with the accuracy rate,F1 score and Io U reaching 97.01%,95.12%and 90.70%,respectively.The model can extract rural residential areas efficiently and accurately,and realize more complete and detailed boundary extraction,with high precision rural residential areas extraction ability.(3)The rural settlements in Liangzhou District is mainly distributed in the central region.From 2010 to 2020,the density and area of rural settlements in this area increased,the cohesion index of residential land gradually increased,the degree of fragmentation decreased,and connectivity increased.New settlements mostly appeared on unused land.The results indicate that the radiation effect of the central urban area in Liangzhou District is constantly increasing,and the spatial distribution of rural settlements is becoming more compact and concentrated.(4)The biggest factor affecting the distribution of rural residential areas in Liangzhou District is soil production potential(0.934),followed by GDP(0.292)and population(0.263)factors.And qualitatively analyzed the impact of natural factors,socio-economic factors,and policy factors on the spatiotemporal evolution of rural residential areas.The results indicate that the changes and development of land use patterns within rural settlements are the result of the combined influence of human activities and many constraints on the natural environment.The spatial distribution of rural settlements is closely related to the production potential of farmland,and is greatly influenced by factors such as GDP,rivers,and roads.The implementation of ecological immigration policies and poverty alleviation policies such as relocation in Liangzhou District has had a significant impact on the spatial distribution and spatiotemporal changes of rural settlements.