高光谱图像地物分类和分割方法研究-模式识别与智能系统专业论文.docxVIP

ABSTRACTABSTRACT ABSTRACT ABSTRACT Hyperspectral remote sensing technique is one current frontier of remote sensing technology．It uses a lot of very narrow band of electromagnetic waves from interesting objects to obtain useful data．Hyperspectral images are acquired by hyperspectral remote sensing imagers and contain rich information including space，radiation and spectral information．However,there exist lots of unresolved problems in theory and applications due to hi．gh—dimension and ambiguity of hyperspectral data．In this thesis， we focus on the classification and segmentation of hyperspectral data．The main work of this thesis is given as follows． Presently,support vector machine(SVM)has been successfully applied to the classification problem of hyperspectral data．Unfortunately,the number of support vectors，non-zero coefficients in the model representation，is too much to result a slow test speed．Since the 1-nornl SVM has better sparser than SVM，we apply the 1-norm SVM to the classification problem of hyperspectral data．The results of experiment verify the effectiveness of the algorithm．The 1-norm SVM Can reduce the number of non-zero coefficients and has a compared performance、析th SVM． Isometric feature mapping(ISOMAP)，a method for dimensionality reduction，has been used to segment landscapes acquired from hyperspectral remote sensing by combining k-means clustering．We find that the similar landscapes cannot be partitioned by using this method．To remedy it，we propose a ISOMAP pixel distributed flow (ISOMAP PD—Flow)method by introducing the space information into ISOMAP．In our method，we connect the feature information谢t11 the space information by a weighted factor．We Can get a space-feature information sequence by changing the values of factor from 0 to 0．5，and use this space-fea觚e information sequence to segment the hyperspectral images．The experimental results show the effectiveness of our algorithm． Although we can segment hyperspectral images by using ISOMAP PD—F