Research On Object-oriented Vegetation Classification Method Based On Texture Feature Of High Resolution Remote Sensing Image

Vegetation is a general term for the plant community covering the surface,and plays an important role in the ecosystem.Vegetation classification is a hot issue in remote sensing application research,very high resolution remote sensing imagery has rich texture information that can effectively improve the classification accuracy.Texture feature extraction is a critical step in remote sensing image processing,the existing texture feature extraction methods ubiquity low accurate classification rate,low efficiency and computing complex.In this study,Naban River Basin in Xishuangbanna,Yunnan Province were used as experimental data.Analysis texture features of vegetation in the VHR imagery,we put forward a kind of vegetation texture feature extraction method based on fingerprint identification technology.In addition,supplemented by texture features for object-oriented vegetation classification,analyzed the effects of texture features on the accuracy of object-oriented vegetation classification.The main research results are as follows:(1)Proposed and implemented a texture feature extraction method based on fingerprint recognition technology.Put forward a texture enhancement algorithm based on fingerprint identification technology,and then combining the fingerprint texture enhancement algorithm with the gray level co-occurrence matrix and the local binary model algorithm,a texture feature extraction method based on fingerprint recognition technology is proposed.This is also the innovation of this paper.Taking the WorldView-2 and Pléiades of the Banban River Basin as experimental data,the proposed texture feature extraction method is used to extract the texture features of the image,then the texture features extracted are used to object-oriented Vegetation Classification.In the WorldView-2 data experiment,compared with the classification result of texture feature based on RGB image extraction,the overall accuracy of object-oriented classification based on fingerprint recognition technology is 91.56%,improved 4.10%,Kappa coefficient is 0.90,improved 0.05.The overall accuracy of object-oriented classification based on fingerprint recognition is 89.36%,improved 3.41%,and Kappa coefficient is 0.87,improved 0.04.In the Pléiades data experiment,compared with the classification result of texture feature based on RGB image extraction,the overall accuracy of object-oriented classification based on fingerprint recognition technology is 88.60%,improved 2.91%,Kappa coefficient is0.86,improved 0.04.The overall accuracy of object-oriented classification based on fingerprint recognition is 84.60%,improved 3.04%,and Kappa coefficient is 0.81,improved 0.04.The results show that the texture feature extracted by the texture feature extraction method based on the fingerprint recognition technology can improve the classification accuracy of the texture rules to a great extent.(2)Add texture feature to object-oriented classification can improve the classification accuracy.In this paper,four texture features extracted by different methods are added to object-oriented classification respectively,compared with the classification method based on spectral information alone.In the WorldView-2 image experiment,the overall classification accuracy of GLCM texture feature based on RGB image extraction is 87.46%,improved 7.05% and Kappa coefficient is 0.85,improved 0.09,the overall classification accuracy of LBP texture feature based on RGB image extraction is 85.95%,improved 5.44%,Kappa coefficient is 0.83,and 0.07 is improved.The overall classification accuracy of the GLCM and LBP texture features extracted based on the fingerprint recognition technique is improved by 11.15% and 8.95%,and the Kappa coefficients are increased by 0.14 and 0.11,respectively.In the Pléiades image test,the accuracy of object-oriented vegetation classification after adding texture features was also significantly improved.The results show that the object-oriented classification method with texture features can improve the classification accuracy of VHR imagery.(3)Multi-classification feature extraction based on single data source for vegetation fine-classification.Based on a single data source,the spectral features,texture features,vegetation index characteristics and geometric features of the image object are extracted and used to object-oriented vegetation classification.The vegetation types are divided into natural forest,rubber forest,Bananas,tea gardens and farmland.In the classification results,the classification accuracy of natural forest is 95.43%,the classification accuracy of rubber forest is 94.33%,the classification accuracy of banana is as high as 93.60%,and the classification accuracy of tea and crop are more than 83.00%.In general,the classification accuracy is good,and realized the vegetation fine-classification based on a single data source.