对宜兴市机载成像光谱数据辐进行射畸变纠正、几何精确纠正和辐射定标等预处理,获得高光谱反射率图像。利用相关系数、均方差、离散度等统计特征进行面向土地利用/土地覆盖目的的波段优选,采用高光谱像元提纯分析工具获得分类所需的终端单元,并利用终端单元进行了研究区土地利用/土地覆盖分类对比研究。研究发现:最大似然法分类精度低,整体精度为84.89%,而二进制编码、神经网络分类和光谱角分类方法精度较高,整体精度分别为87.12%,88.75%,90.41%。这说明光谱角分类是最有效的高光谱影像土地利用/土地覆盖分类方法,文章研究采用的高光谱预处理以及波段选择方法是正确的。

Taking Yixing City in Jiangsu Province as an example, a map of reflectivity is acquired after serials of pre-processing steps such as the calibration of illumination aberrance, rectification of geometry and calibration of reflectivity have been done. The effective bands are determined by analyzing the parameters of correlation of bands, entropy and separability of 128 bands. Based on the analyzing tools of hyper-spectrum images, some endmembers are acquired, and a comparison study of classification of land use/land cover has been done. The study shows that the overall accuracy of maximum likelihood is about 84.89%, and Kappa Coefficient =0.74, while overall accuracy of binary encoding, neural net and spectral angle mapping is 87.12%,88.75%,90.41% respectively. It also shows that the processing step is correct in the study, and spectral angle mapping is the most effective for imaging spectrometer data classification of land use and land cover.