HSV变换与监督分类结合的沙漠信息提取  

Desert information extraction using HSV transformation combined with supervised classification

作　　者：俞天琦 郝媛媛 白小明[1] YU Tianqi;HAO Yuanyuan;BAI Xiaoming(Pratacultural College of Gansu Agricultural University/Key Laboratory of Pratacultural Ecosystem of Ministry of Education/Engineering and Technology Research Center for Alpine Rodent Pest Control,National Forestry and Grassland Administration,Lanzhou 730070;Hulunber Grassland Ecosystem National Observation and Research Station/Institute of Agricultural Resources and Regional Planning,Chinese Academy of Agricultural Sciences,Beijing 100081,China)

机构地区：[1]甘肃农业大学草业学院/草业生态系统教育部重点实验室/国家林业草原高寒草地鼠害防控工程技术研究中心,兰州730070 [2]中国农业科学院农业资源与农业区划研究所/内蒙古呼伦贝尔草原生态系统国家野外科学观测研究站,北京100081

出　　处：《干旱区资源与环境》2023年第12期103-114,共12页Journal of Arid Land Resources and Environment

基　　金：甘肃省大学生创新创业训练计划项目(s202010733013);甘肃农业大学大学生创新创业计划项目(202019009);国家自然科学基金项目(41907406);甘肃农业大学科技创新基金项目(GAU-KYQD-2018-23)资助。

摘　　要：沙漠化是全球干旱、半干旱地区土地退化的主要表现之一,而如何快速、高效、准确、大范围的提取沙漠信息一直是沙漠化研究领域的技术难点。文中以石羊河下游民勤盆地为研究区,以Landsat-8卫星影像为数据源,采用HSV色彩变换的光谱增强手段,探讨了HSV变换对沙质地表像元的影响,对比分析了6种监督分类方法(神经网络、支持向量、最大似然、马氏距离、平行六面体和最小距离)对于沙漠信息提取的效果和精度。结果表明:1)HSV变换降低了沙质地表像元在Red和NIR波段的辐射亮度值(分别为增强前的1.0%和29.8%)及其标准偏差(0.81%和19.7%),提高了SWIR 2波段的辐射亮度值(1.50倍)及其标准偏差(1.32倍);提高了沙质地表与其他土地之间的分离程度,增大了沙质地表与其他地表在光谱层面的差异。2)最小距离法在HSV变换前后均具有较高的精度(Kappa系数分别为0.9615和0.9627),且由于其方法简单、运算时间短,在研究区沙漠监测中具有较好的应用前景;HSV变换明显提高了平行六面体与最大似然法的精度,Kappa系数分别增加了0.5400和0.2048;HSV变换与神经网络法相结合是快速准确提取沙质地表信息的最优方法,Kappa系数可达0.9682。以上结论可以为今后沙漠信息提取与监测提供理论依据。Desertification is one of the main manifestations of land degradation in arid and semi-arid areas of the world,and how to extract desert information quickly,efficiently,accurately and extensively has been a technical difficulty in the field of desertification research.In this study,the Minqin Basin in the lower reaches of Shiyang River was taken as the research area,Landsat-8 satellite image was used as the data source,and the spectral enhancement method of HSV color transformation was used to discuss the influence of spectral enhancement on the sandy surface pixels.The effects and accuracy of six supervised classification methods(neural network,support vector,maximum likelihood,Mahalanobis distance,parallelepiped and minimum distance)on desert information extraction were compared and analyzed.The results show that:1)the spectral enhancement decreases the radiation luminance value(1.0%and 29.8%before enhancement)and the standard deviation(0.81%and 19.7%)of the sandy surface pixels in Red and NIR bands,and increases the radiation luminance value(1.50 times)and the standard deviation(1.32 times)of SWIR 2 bands.The separation degree between the sandy surface and other land is improved,and the difference between the sandy surface and other land is increased at the spectral level.2)The minimum distance method has high accuracy before and after spectral enhancement(Kappa coefficient is 0.9615 and 0.9627,respectively),and has a good application prospect in desert monitoring in the study area due to its simple method and short operation time.The spectral enhancement significantly improves the accuracy of parallelepiped and maximum likelihood methods,and the Kappa coefficients increase by 0.5400 and 0.2048,respectively.The combination of spectral enhancement and neural network method is the best method to quickly and accurately extract sandy surface information,and the Kappa coefficient can reach 0.9682.The above conclusions can provide theoretical basis for desert information extraction and monitoring in the future.

关 键 词：HSV变换 监督分类 沙漠信息提取 民勤盆地 干旱、半干旱区 

分 类 号：X24[环境科学与工程—环境科学]