As the quad-polarimetric SAR data has plentiful information, it is difficult to obtain good classification results just by single polarimetric feature and pixel-based classification. Based on the X-band quad-polarimetric radar data of Terra SAR-X in Mengla of Xishuangbanna and Simao Pu'er city, Yunnan province, the object-oriented land cover classification experiments based on feature selection are carried out. Firstly, the quad-polarimetric SAR data is preprocessed. The Pauli RGB image is extracted from the study area data, and then is segmented as the basic unit of classification using the image segmentation technique. Then, the polarimetric features and the texture features of Span images are extracted from the SAR images, and the optimal feature sets are selected. Finally, the object-oriented fuzzy classification method is used into classification experiments, and the classification results are evaluated by field survey data. The experimental results show that object-oriented method can effectively remove the impact of noise, and the optimal combination of feature bands makes the classification results more accurate. The overall classification accuracy is up to 88.5% for Mengla in Xishuangbanna, and is 86.8% in Simao of Pu'er city. Compared with the H/A/alpha -Wishart classification method, the accuracy is improved by more than 40%.%全极化SAR数据信息丰富,仅利用单一的极化特征和基于像元的分类很难得到较好的分类效果.因此,提出了全极化数据特征优选结合面向对象方法进行土地覆盖分类.以云南西双版纳州勐腊县和普洱市思茅区的Terra SAR-X的X波段全极化雷达数据为信息源,首先对全极化SAR数据进行预处理,提取研究区Pauli RGB图像后,利用影像分割技术对Pauli RGB图像进行分割,作为分类的基本单元;然后对SAR影像提取极化分解特征和Span影像的纹理特征,选取最优特征集合;最后利用面向对象模糊分类方法进行土地覆盖分类,并采用实地调查数据对分类结果进行了精度评价.试验结果表明,面向对象方法可以很好地去除噪声的影响,最优组合的特征波段使得分类结果更加精确.西双版纳州勐腊县总体分类精度达到88.5%,普洱市思茅区总体分类精度达到86.8%,较之H/A/α-Wishart分类方法精度提高了40%以上.
展开▼
