Research On Farmland Boundary Extraction Algorithm From Remote Sensing Images Based On Deep Learning And Morphology

Agriculture occupies a pivotal position in China,and with the advancement of modern agriculture,it is imperative to build information-based farmland and smart farmland.Information on the location and extent of farmland units is necessary for many applications of smart farmland,so the problem of extracting precise automated farmland boundaries is in urgent need of solution.With the continuous development of remote sensing technology,it is possible to obtain large-scale farmland information,especially high-resolution remote sensing satellite data,which can capture more feature details and texture information and provide a good data basis for farmland boundary extraction.Based on the high-resolution GF-2 satellite remote sensing images,this thesis carries out relevant research around the problem of automatic farmland boundary extraction,based on the combination of deep learning and morphology to realize the accurate extraction of farmland boundaries,and designs a farmland boundary extraction system.The specific research and work are as follows.(1)Research on farmland boundary extraction algorithm based on deep learning and morphology.In this thesis,deep learning algorithm is combined with morphological edge detection.Firstly,this thesis compared the performance of three classical deep learning algorithms,UNet algorithm,HED algorithm and CNN algorithm,on farmland boundary segmentation task.Secondly,this thesis used Canny operator,SUSAN operator and Sobel operator for farmland boundary extraction of farmland images respectively.Finally,combined with the theory of deep learning and morphology,a farmland boundary extraction algorithm based on UNet network and Sobel operator is proposed,and the farmland boundary is extracted with high precision.The experimental results show that the structure similarity SSIM of the proposed method reaches 98.88%.(2)Farmland boundary segmentation algorithm using remote sensing images based on MDE-UNet.To achieve high precision segmentation of farmland boundary,a Multi-task Deformable UNet combined Enhanced network(MDE-UNet)is proposed for farmland boundary segmentation.The network consists of two parts: a Multi-task Deformable UNet(MD-UNet)segmentation module with Deformable UNet(D-UNet)as the basic network and an enhancement module with a lightweight UNet improved by residual attention.In the MD-UNet segmentation module,three branches are used for precise segmentation of deterministic,fuzzy,and raw boundary,respectively.In the basic network of MD-UNet segmentation module,a weight-modulated deformable UNet(D-UNet)network is used to improve the directional perception capability of this network.In the enhancement module,the lightweight UNet improved by residual attention is cascaded with the MD-UNet segmentation module,which could further enhance the segmentation accuracy of the MD-UNet segmentation module fusion results.The experimental results show that the F1 score and Kappa coefficients of the farmland boundary segmentation results obtained by the MDE-UNet model proposed reach 94.49% and 92.23%,respectively.Compared with the UNet model,the overall accuracy of this model has been greatly improved,and the F1 score and Kappa coefficient have been increased by 8.19% and 12.70%,respectively.(3)The design of farmland boundary extraction system based on remote sensing images.Based on the Qt application development framework,the farmland boundary extraction system is designed.By using the proposed MDE-UNet model,together with morphological extraction and skeleton extraction algorithms,the system can realize the segmentation and extraction of farmland boundaries and farmland boundary centerline extraction functions,as well as the functions of accuracy evaluation of farmland boundary segmentation images and area estimation of farmland.In addition,the parallelized processing strategy of remote sensing images,which means parallelized cropping and reconstruction of segmentation results,is designed to realize the segmentation and extraction of farmland boundaries for remote sensing images of larger areas.To achieve high precision segmentation of farmland boundary,a farmland boundary extraction method combining UNet network and Sobel operator is used.The MDE-UNet model proposed has realized the high-precision segmentation of farmland boundaries.In addition,the farmland boundary extraction system is designed to implement farmland boundary segmentation,extraction and farmland boundary centerline extraction functions with excellent farmland boundary extraction accuracy,which can meet the application requirements of smart agriculture.