Object-Oriented Landsat8 Remote Sensing Images For Desertification Land Classification

Land desertification is caused by the uncoordinated development of human and nature in recent centuries.In order to monitor desertification and prevent desertification,a research area based on CART decision tree classification and object-oriented desertification is proposed,focusing on the Shapotou District of Zhongwei City,Ningxia.The land extraction algorithm,through theoretical analysis and experimental research,proposes four ways to improve the classification accuracy,optimizes the sand extraction feature information,and realizes desertification land classification and information extraction.Remote sensing data has the characteristics of high space,high spectrum and high temporal resolution.In the classification of desertification and classification of desertification,combined with the situation of the research area,refer to the National Technical Regulations on Desertification and Desertification issued by the State Forestry Administration in 2004.The adaptability of high spatial resolution and moderate time resolution of the eighth generation remote sensing satellite Landsat8 to desertified land information was selected.The remote sensing data of Shapotou area in Zhongwei was selected and subjected to atmospheric correction and aerosol inversion.In the process of training CART decision tree rules,firstly,according to the desertification indication factor and previous research results,four characteristics of 24 spectra,customization,texture and slope for extracting desertification land information were selected,and MSAVI and SDI were corrected.The calculation of the two features achieves the inversion of all objects.Then through the object-oriented thinking,multi-scale segmentation and spectral difference segmentation algorithm are used to segment the remote sensing image to obtain the data of the object level,avoiding the "salt and salt phenomenon" produced by the homologous and the same-spectrum foreign matter;The representative and complete sample points are obtained;finally,the selected features and sample points are input into the CART decision tree trainer to generate a classification rule tree.Using object-oriented thinking and CART decision tree classification method,18 characteristics for desertification classification and extraction were learned.According to this rule tree,desertification classification experiments were carried out in remote sensing images,and the results were analyzed and analyzed by confusion matrix.For 82.3%,the Kappa coefficient is 0.796,which indicates that the algorithm can effectively extract desertification information.In order to improve the classification accuracy,four improved methods of multi-source remote sensing image fusion,selection of appropriate segmentation scale,multi-feature participation extraction and selection of appropriate classifiers are proposed.Through the research of this paper,the following conclusions are obtained:Landsat8 data is remote sensing image data suitable for desertification information extraction application;the classification result of extracting desertified land is high and has practical application value;by integrating multiple data sources,optimizing segmentation scale,multi-feature participation,The selection of excellent classifiers can improve the classification accuracy;desertified land is mostly distributed in the north of the Yellow River in Shapotou District,and there is little light desertified land covering the south of the Yellow River.The desertification process changes from extremely severe to slightly outward.