Identification Of Primary Land Cover Types Of Hefei City Based On Time Series Landsat Imagery

With the rapid development of social economy,the spatial pattern of land cover has changed dramatically.Land cover has always been one of the key elements in the study of land cover information.The most important prerequisite in land cover is land cover classification,in order to develop and utilize land resources better.Classification and extraction of land cover information is very necessary.Traditional artificial visual interpretation and pixel classification can not meet the rapid development of urban and remote sensing image development,so object-oriented image extraction came into being.In this thesis,the land cover classification of Hefei city is studied by using the Landsat-5 TM data of 2000,2005 and 2010 and Landsat-8 OLI data of 2010 as the data source.Feature selection and image classification are used to improve the classification accuracy and efficiency of land cover classification.The main contents and conclusions are as follows:(1)Considering the image quality,multiple phases and the geographical environment of the study area,Finally,the Landsat-5 TM images of 2000,2005 and 2010 of Hefei and Landsat-8 OLI remote sensing images of several phases are selected as the research data.Data preprocessing is the first and key step.Geometric correction,image cutting and other preprocessing,finally get Hefei image data.(2)In this thesis,four segmentation algorithms of eCognition Developer 9.0 are analyzed in detail and compared with each other under the same parameters.In addition,the segmentation results of different segmentation scales are also tested,and the segmentation of different ground objects in different pixels is taken as the standard.The best segmentation method is to use multi-scale "over-segmentation"first,and then to optimize the segmentation by spectral difference segmentation by means of segmentation experiment,visual interpretation and accuracy evaluation of confusion matrix.(3)In this thesis,two key technologies for image information extraction in eCognition Developer 9.0,nearest neighbor supervised classification and SEaTH algorithm rule classification,are described,and their advantages and disadvantages are analyzed.Although the nearest neighbor supervised classification method is simple,it needs to calculate the central distance between the objects to be classified and the selected samples for each feature classification.Especially when the texture feature is involved,the algorithm can automatically determine the optimal feature and the best threshold,but when the JM distance between the objects is poor,the order of the ground object classification can not be determined.Using the SEaTH algorithm and the nearest neighbor supervised classification,land cover is classified into six categories:woodland,grassland,cultivated land,wetland,artificial surface and other six categories.The classification is compared with the results of the remote sensing survey and assessment of the ten-year change of the ecological environment in China,and the accuracy of the classification is verified.The results show that the total accuracy and Kappa coefficient of land cover classification are increased to more than 90%by the combination of SEaTH algorithm and the nearest neighbor supervised classification.