Land Cover Classification Of Remote Sensing Images Based On Expert Knowledge And Improved SEaTH

The study of land cover change is an important part of China's current geographical situation monitoring,and it is also the basis for many special studies in global change research.Land cover classification is the premise of land cover change research.At present,the classification of land cover is mainly through the analysis and interpretation of remote sensing images.The land cover classification accuracy based on remote sensing images directly determines the level of land cover research.Improving classification accuracy is a hot issue in land cover research.Land cover classification methods include computer automatic classification and expert knowledge manual classification.When classifying complex surface land cover distribution,the direct application of the computer is due to the influence of mixed pixels,staggered distribution of land types,and land cover crushing.The accuracy of classification results obtained by automatic classification is low,and the subjective factors of knowledge classification of applied geography experts are influential and scientifically inadequate for classification,and a large amount of manual operations are required.Therefore,the overall global classification accuracy of land cover data is low.This paper analyzes the disadvantages of using expert knowledge classification and computer automatic classification in land cover classification.The accuracy of artificial classification against expert knowledge is high,but it requires a lot of manual operation,long interpretation cycle,and classification results are subject to Subjective factors have a greater impact.The classification of computerized automatic classification methods takes a short time,but the classification accuracy is low when the land cover distribution is complicated.This paper combines the two,proposes an improved Separation Thresholds algorithm(SEaTH,Separability and Thresholds),and calculates an optimal threshold based on Gaussian probability.This method improves the SEaTH algorithm as a semi-automatic algorithm.It can only calculate the classification threshold of every two classes,and automatically filters the optimal classification rules and classification thresholds by computer to achieve semi-automatic SEaTH algorithm automation.It is difficult to classify land cover land,and the red edge band of vegetation is added into the classification to explore the role of red edge band of vegetation in land cover classification of remote sensing images.Compared with previous studies,this article has achieved the following three results:(1)Under the premise that the feature space accords with the Gaussian probability distribution,based on the improved SEaTH algorithm and the expert knowledge classification result,compared with the machine learning classifier,the classification accuracy is equivalent,and the classification effect obtained in some specific land cover categories is better than Machine learning classifier classification effect.(2)The classification of forest land is a difficult point in the land cover classification.The vegetation red edge band is used to study the classification of forest land in the classification of land cover,and the sensitivity of the red edge band of vegetation to the change of vegetation chlorophyll is used to add vegetation red.Edge bands improve the classification accuracy of land cover land classification.(3)The land cover classification method based on expert knowledge and improved SEaTH algorithm was applied to the experiment in the study area.The overall accuracy was 72.96%.Compared with the K-NN algorithm,the nearest neighbor pixel method,and the CARTS algorithm,the classification accuracy is increased by at least 5.35%.Compared with the random forest algorithm,the overall accuracy of the classification is improved by 3.64%.