Optimizing Image Segmentation Based On High-precision Land Cover Data Method And Application In Natural Resource Monitoring

Natural resources are the foundation of development,the source of ecology and the basis of people's livelihood,which play an important support and guarantee role for economic and social development.With the increasing population of our country and the acceleration of urbanization,it is of great significance to grasp the quantity and spatial distribution of natural resources to cope with the increasing demand for natural resources,to do a good job in the supervision of natural resources and land and space planning.It is therefore imperative to monitor natural resources in a timely and accurate manner.At present,the third national land survey and other natural resources monitoring mode to manual visual interpretation,low degree of automation.Image segmentation is an important basis for solving the automation of natural resource monitoring,but the current method of image segmentation only considers the inherent characteristics of the primitive's own spectrum,shape and texture,neglects the analysis and application of the boundary characteristics of the primitive,and it leads to the phenomenon of low boundary positioning accuracy,too fine fragmentation results and poor overallity.In this paper,the boundary prior knowledge constraint of high-precision land cover data is integrated into multi-scale segmentation,optimizing the high-resolution image segmentation results to solve the problems of low boundary accuracy of split objects,so as to improve the accuracy and efficiency of natural resource monitoring.On the basis of optimizing the results of segmentation,this paper explores the application of multi-feature variable extraction and optimization and classifier algorithm in land-use classification,as well as realizes the application of multi-source and multi-sequence data in the automatic extraction of information of returning farmland to forest.The results of the study show that:?1?The high-precision land cover data optimizating image segmentation method makes the boundary between image objects more accurate.Based on fractal network evolution algorithm,the segmentation method integrates the boundary constraints of the a priori results of land cover data into multi-scale segmentation,and obtains the optimal segmentation scale parameters combined with ESP scale evaluation tools and normalized maximum Minimum area index segmentation evaluation.This method solves the problem of low precision of dividing object boundary,and the boundary of the object after treatment is more consistent with the outline of the real object,and the segmentation effect is better.?2?The high-precision land cover data optimizating image segmentation method combined with the multi-feature variable optimization method greatly improves the precision of land use classification.Based on the high-precision land cover data optimization segmentation method,this study uses the classification method based on high-precision land cover data optimization segmentation and the non-land cover data auxiliary classification method,and uses simple Bayes,decision tree,random forest and the nearest neighbor classifier to carry out the land use information extraction in Wugong County.The classification method of land cover data optimization image segmentation boundary has greatly improved in accuracy,its general accuracy is 95.3%,Kappa coefficient is 0.94,which showed that this method has good feasibility and validity for extracting land-use information.?3?The land cover data optimizating image segmentation method combined with the time-series characteristics of Sentinel-2 image greatly improve the precision of returning to the forest.In this study,combined with the boundary characteristics of high-resolution imaging and the weather characteristics of Sentinel-2,the experiment spree shows that the most effective way to distinguish the retreating farmland forest and other land classes is the characteristic difference between the Red Edge Index?NDVIre2+NDVIre3?in November and October,the membership degree classification based on?NDVIre2+NDVIre3?_11-10 is more accurate than the other three methods.The precision of information extraction of returning farmland to forest in Wangyi and Wangyi country is over 90%,which shows that the method is feasible and applicable for the extraction of the return of forest information.