【作者】 李小梅；

【导师】 张秋良；

【作者基本信息】 内蒙古农业大学 ， 森林经理学， 2010， 硕士

【摘要】 目前多角度高光谱遥感技术成功应用到了多种学科中,取得了一些研究成果,并展示了其应用潜力。在林业中,国外展开了植被叶面积指数、生物量、植被生化信息、针叶树种识别等方面的研究工作。我国最近两年也在不断的挖掘及开展多角度高光谱遥感在林业中的应用研究。本文以吉林省长白山为试验区,针对CHRIS/PROBA多角度高光谱遥感数据的特点,对其进行预处理的基础上,首先通过应用最大似然法(MLC)、最小距离法、支持向量机法(SVM)和光谱角填图法(SAM)等几种常用的遥感影像分类方法对CHRIS 0°影像进行森林类型分类,结果进行比较分析获取精度较高的分类方法;然后采用精度最高的分类方法对CHRIS五个角度影像进行分类,结果进行比较分析;之后又采用该方法对多角度组合影像进行分类,结果分析评价;最后波段组合影像采用该分类方法进行分类,结果分析评价。主要研究结论:采用不同的分类方法对CHRIS 0°进行分类,结果分析比较得出:SVM分类精度最高72.8448%,Kappa系数为0.6770;利用SVM对CHRIS五个角度影像进行分类,结果比较分析得出:分类精度从高到低依次排为FZA=0>FZA=-36>FZA=-55>FZA=36> FZA=55;利用SVM对多角度组合影像进行分类,结果与CHRIS 0°结果分析比较得出:多角度组合影像森林类型总体分类精度低于单角度影像分类精度;利用SVM对波段组合影像进行分类,结果与CHRIS 0°比较分析得出:波段组合影像森林类型总体分类精度很低,效果不如多角度组合分类结果。本研究得出长白山自然保护区森林类型分类研究中采用CHRIS多角度高光谱数据进行分类,应采用SVM方法对CHRIS单角度影像进行分类效果最好。更多还原

【Abstract】 Multi-angle hyper-spectral remote sensing had been applied successfully in various subjects, and gained some research achievements. The application potential had been showed. In the field of forestry, research works such as LAI, biomass, vegetation biochemical information, coniferous trees identification and so on are under way abroad. Application research of multi-angle hyper-spectral remote sensing in forestry had been carried continuously recent years in China too.In this paper, Changbai Mountain is selected as study area. According to the characteristic of CHRIS/PROBA multi-angle hyper-spectral remote sensing data, several common remote sensing image processing methods such as Maximum Likelihood Method (MLM), Minimum Distance Method (MDM), Support Vector Machine (SVM) and Spectrum Angle Mapping (SAM) are applied to classify forest types in CHRIS 0-degree image based on pre-processing. Classification method with relatively higher precision is acquired by results comparison. Then the method is applied to classify five angle images of CHRIS and the results are compared and analyzed. Afterward, the method is applied to classify multi-angle combination images and the results are compared and analyzed. Finally, the method is applied to classify band combination images and the results are compared and analyzed.Main study conclusion:By comparing and analyzing classification results of different method to CHRIS 0-degree image, SVM has the highest classification precision of 72.8448%, while the Kappa index is 0.6770. By comparing and analyzing classification results of SVM to CHRIS 5 angle images, the classification precisions are sorted as follow: FZA=0>FZA=-36>FZA=-55>FZA=36>FZA=55. By comparing and analyzing the classification results of SVM to multi-angle combination images with CHRIS 0-degree image, the overall tree type classification precision of multi-angle combination is lower than single-angle image. By comparing and analyzing the classification results of SVM to band combination images with CHRIS 0-degree image, the overall tree type classification precision of band combination is very low which is lower than multi-angle combination image classification. It is concluded that, application SVM to CHRIS single-angle image is the best method to classify forest type by CHRIS multi-angle hyper-spectral data in Changbai Mountain natural reserve.更多还原