基于机载CCD和ALS伪波形数据的山区地表分类研究

为利用机载激光雷达(Airborne Li DAR scanning,ALS)、结合高空间分辨率影像进行土地利用分类,提出一种利用统计高程分布曲线生成的ALS伪波形,结合点云强度信息和CCD影像RGB 3波段数据对山区复杂地表进行分类的方法,并验证了该方法对山区复杂地形下典型地物的分类精度。通过安徽黄山地区研究区数据分类结果与相同区域基于光学图像的Globe Land30全球分类产品的对比,验证了该分类方法的可行性和适用性。利用伪波形结合强度信息和RGB信息生成的分类特征曲线,采用神经网络分类方法(ANN)将研究区内地物分为农田、森林、水体、村庄4类。结果表明,研究区分类总体精度达到95. 22%,Kappa系数0. 919 2,较同一区域、同等分辨率的光学数据分类产品(总体精度79. 56%,Kappa系数0. 661 8)精度显著提高。

Land Cover Classification in Mountainous Area Based on Airborne CCD Image with LiDAR

In order to fuse airborne LiDAR scanning (ALS) and high resolution image for land cover classification, modeled LiDAR waveform that based on elevation information was integrated with CCD image three band (RGB) data in complex mountainous area. Verification experiment results showed that high accuracy for several typical land cover types classification in complex mountainous area could be acquired by the method. Moreover, the classification result of study area which located in Huangshan, Anhui Province was used to compare with the GlobeLand30 classification result to verify the method. The entire coverage of study area were classified into four land cover types (farmland, forest, water and village) by characteristic curves that combined modeled waveform with intensity and RGB information through artificial neural network (ANN). The result showed that the overall accuracy of study area classification was 95.22%, and Kappa coefficient was 0.9192. Compared with GlobeLand30 classification result accuracy (overall accuracy was 79.56% with Kappa coefficient of 0.6618) in this area, the research results was improved significantly.