地空同步试验高光谱影像特征提取与分类

Feature extraction and classification of hyperspectral image with ground-sky synchronization test

针对地空同步的高光谱特征提取与反演理论研究匮乏的现状,结合“类内密度最大,类间距离最大”原则,探索不同谱段对不同地物识别的可分性和重要性选择方法,并提出改进的投影寻踪分类方法,实现不同地物在特征波谱的加权投影寻踪。以山西省临汾市某苗圃为实验区,同步采集实验区不同地物光谱与PHI高光谱影像,依托实测光谱,围绕先整体最优后局部最优策略,构建不同分类规则,并将其应用于PHI高光谱影像分类,实现植被与非植被的信息提取,并细分为10余种不同地类。结果表明,420nm,520nm,570nm,610nm,660nm,690nm,715nm及810nm等8个谱段为不同植被的重要可分性谱段,分类结果具有层次感强、植被轮廓清晰、避免阴影影响等优点。

In view of the lack of theoretical research on the extraction and inversion of hyperspectral features in ground-sky synchronization, the authors, in combination with the principle of “maximum density within class and maximum distance between classes”, studied the separability and importance selection of the spectra for different ground objects in different spectral regions, proposed an improved projection pursuit classification method, and realized the projection pursuit method based on weighted feature band. In the case study, the spectral and PHI hyperspectral images of different ground objects in the experimental area were collected synchronously and, combined with the measured spectra on the ground, constructed different classification rules around the strategy of overall optimization and local optimization. It was applied to the classification of PHI hyperspectral image to extract the information of vegetation and non-vegetation and to subdivide more than ten different kinds of ground objects. The results show that 8 spectral areas are important separability bands of different vegetation: 420 nm, 520 nm, 570 nm , 610 nm, 660 nm, 690 nm, 715 nm, and 810 nm. The classification results have the advantages of strong stratification, clear outline of vegetation and avoiding shadow influence.