Research Of Land Classification From Remote Sensing Images Based On Object-oriented With Texture Feature

The rational use of land resources related to national economic development and nationalsurvival, so supervised the change of land use and land cover is an important content to realizethe reasonable utilization of land. The development of high spatial resolution image, makingimages displayed more special information and texture information and context information andso on, and promoted the research of land use and land cover change detection. Researched onland use situation and the dynamic change of land use with high spatial resolution remote sensingtechnology, provided scientific basis for the departments of land management in land useplanning and land management. Need to extract the changes of land use during different periodsfirstly while supervised the change of land use, the method of after classification comparison wasoften used to extract the land use change. The classification based on pixels can only use of thespectral information of remote sensing data, not only reduce the accuracy of classification, butalso waste a lot of information resources, while the technology of object-oriented classificationbased on objects can make full use of the abundant information of high spatial resolution image,improved the accuracy of classification, and the defects of the classification based on pixels aresolved from a certain extent.This article researched on the classification of image was based on the technology ofobject-oriented classification with texture features. Choosing Gangzha district in Nantong as theresearch region, and obtained the information of land use from two images of2008and2012,then extracted the change of land use and land cover for each type in four years from2008to2012through the classification results. Following are the main research contents andachievements in this paper:(1) Inductive and analysis of the research status for present stage of supervise for land usechange at home and abroad, according to the characteristics and problems of the existed researchmethods, using the method of after classification comparison to extract the change of land usebased on object-oriented with texture feature. Classified the images based on object-oriented withtexture features and that without, and do visual judgment and accuracy analysis of both twoclassification results，reached that the accuracy of object-oriented classification is higher.(2) Preprocessing the images of research region in2008and2012. In order to make thebands are used to fusion containing the greatest amount of information, to calculate the standarddeviation of each band, covariance matrix and correlation coefficient matrix between two bandsexcept panchromatic band, and based on the calculation reached the best combination of threebands.(3) Using gray level co-occurrence matrix to extract image texture features of differentwindows, according to the trends analysis model for texture characteristic value changing with the different size of window, reached the best window size to extract texture features. Otherwiseextracted the texture index images for image data based on ENVI, and put forward the effectrelationship of texture feature to do statistics and analysis for the texture index images on thebasis of gray level co-occurrence matrix can used to statistic the number of pixels occurrencewhich have same value in the area of image, based on the result chose the best combination oftexture features which are joined into object-oriented to classified images.(4) Using the method of maximum number of complete objects to choose the bestsegmentation scale from the results that have been segment by multi-scale, when the model wasused to calculate the best segmentation scale we need to set a threshold through experiments, andfinally reached the best segmentation scale for rural road and plough and water and constructionand unused and highway and woodland of research region is respectively80and100and150and180and140and90and170.(5) Analysis of dynamic change of land use from2008to2012, through the analysis modelfor the changes of land use quantity and land use structure and land use degree, reached that largechanges had taken place in the land utilization condition of the study area in the past four years,the trends of the area growth of plough and unused and highway are obviously, the areas ofconstruction and water and woodland and rural road are reduced. The rapid growth of unusedland areas made the level of land use in2012become lower. Finally, use Markov predictionmodel to predict and analysis the state of land use in the future for eight years.