Research On Farmland Segmentation Technology Based On High Resolution Satellite Images

Arable land is one of the important resources for human survival,which can ensure the supply of food and agricultural products and provide essential support for national economic development and food security.The 20 th National Congress of the Communist Party of China emphasized the importance of strengthening the foundation of food security,strictly protecting the "red line" of 1.8 billion mu of arable land,and establishing a benefit compensation mechanism for major production areas,to ensure that the Chinese people can firmly hold their rice bowls in their own hands.High-resolution satellite images have the characteristics of high resolution,wide observation range,objective data,and strong real-time performance,which save a lot of manpower and material resources in satellite image processing.Therefore,satellite images have achieved great success in semantic segmentation,object detection,change detection,and other fields.Their research results have been widely applied in the agricultural field,especially in arable land protection,where high-resolution satellite images can assist in plot demarcation,land resource management,modernization of agriculture,and formulation of arable land protection policies,and have broad application prospects.Currently,deep learningbased segmentation algorithms in satellite image processing suffer from problems such as inaccurate target instance recognition,difficult processing of complex textures,and weak generalization ability.Meanwhile,most satellite image datasets have issues such as uneven sampling of instance,small scale,insufficient refinement of class categories,and inaccurate sample labeling.To address these issues,this dissertation comprehensively utilizes satellite images,deep learning-based semantic segmentation algorithms,and geographic information systems,and combines them with practical application scenarios to construct a satellite image dataset based on domestic high-resolution satellite images.Then,this dissertation proposes an arable land segmentation algorithm based on satellite image texture features and edge features,and finally develops a web-based GIS arable land segmentation technology application system.The main research contents of this dissertation are as follows:(1)Construction of arable land dataset based on high-resolution satellite imagery.In this dissertation,the multispectral image and panchromatic image are fused through the preprocessing of GF-2 multispectral and panchromatic satellite images.The multispectral images are fused with the panchromatic image to produce images with a resolution of 0.8 meters.We select the area of interest for segmentation and take this as a basis for constructing the UJNLand dataset of high-resolution satellite imagery.This dataset includes wheat,roads,bare soil,water bodies,and other background types,with 700,000 sample images,which is the largest dataset for arable land segmentation in China and internationally.(2)Research on arable land segmentation algorithm based on texture features.This dissertation proposes a texture feature extraction network based on Gabor filters.This network is based on the U-Net network and uses multiple groups of Gabor filters with different directions and scales to extract texture features of the image.The original image is first processed by the Gabor filter for texture feature extraction,and a single-channel image of the texture feature is obtained before it is input into the U-Net network along with the original three-channel image.An SE block is added before upsampling,and an adaptive structural optimization and residual connection method is used to improve the network’s feature extraction capabilities and robustness.(3)Research on arable land segmentation algorithm based on edge features.Based on Seg Net,this dissertation proposes a Laplacian filter embedded attention network and adds a Laplacian filter and convolutional block attention module to extract image features further and aggregate the context information of different regions based on the Seg Net encoder.Experimental results demonstrate the effectiveness of the proposed method on our dataset.(4)Application system of arable land segmentation technology based on Web GIS.This system combines remote sensing technology,image segmentation algorithms,and Web GIS technology to segment satellite imagery,extract arable land information,and display it visually through Web GIS technology.The system provides support for satellite imagery and related applications by offering user management,category management,query,image download,dataset display,arable land segmentation,and download logs...