【摘要】 将机载CASI和SASI高光谱数据整合,既可以获取可见光—近红外—短波红外区间连续的窄波段地物光谱,又能得到很高的空间分辨率,为高覆盖度的森林树种识别又增加了一种新方法。但是由于两种传感器的光谱响应不同,接收到的辐射值差异较大,如何将两种数据有效整合目前仍是一个难题。CASI和SASI覆盖谱段不同,受大气影响程度也不同,根据植被反射和吸收光谱特性,首先用基于统计模型的经验线性法和基于辐射传输的MODTRAN模型分别对CASI和SASI大气校正,复原地物光谱真实的反射率。然后去除反射率光谱包络线,用Savitzky-Golay滤波函数对归一化后的光谱曲线进行平滑,以去除噪声及异常点,实现CASI和SASI数据(CASI+SASI)的整合。与实测光谱曲线对比发现,整合后的CASI+SASI光谱曲线与实测光谱曲线匹配度较高,并且比单一传感器的光谱信息更丰富,有利于不同树种的区分识别。最后应用光谱微分及曲线匹配技术,选取SVM分类器实现了研究区的树种填图,总体精度达到86.21%,Kappa系数为0.8297,该方法有效可行,为后续的相关研究提供了参考。更多还原

【Abstract】 Integrating CASI and SASI airborne hyperspectral data could help acquire both continuous narrow-band spectra covered the visible-near infrared-shortwave infrared range and high spatial resolution information.It provides a new method for tree species identification in high coverage rate forest.However,due to the different spectral response of the two sensors,there were differences between the received radiances.How to integrate the two dataset effectively for tree species classification is still a problem.CASI and SASI data have both different coverage of spectrum and atmosphere impact.According to the characteristic of reflectance and absorption spectra of vegetation,experience linear method based on statistical model and MODTRAN model based on radiative transfer theory were used for atmospheric correction of CASI and SASI data respectively to retrieve the real spectral reflectance of ground object.Then CASI and SASI data were integrated(CASI + SASI) by Savitzky-Golay filter function smoothed continuue removal spectra to reduce noise and abnormal points.Compared the integration of CASI + SASI spectrum with the field measured spectrum showed that the two curves matched well each other.In addition,the integrated spectra have richer spectral information for tree species distinction than any single one.Tree species mapping was realized base on the integration of CASI and SASI data using SVM classifier.An overall accuracy of 86.21% and Kappa coefficient of 0.8297 were obtained.And the result indicated that the proposed method to integrate CASI and SASI data is feasible for a more accuracy in forest tree species classification,and it would be a reference for the later research.更多还原