【摘要】 利用高分一号影像结合机载LiDAR数据进行面向对象的亚热带森林年度采伐迹地分类提取。在面向对象的遥感软件Ecognition中,首先利用森林小班数据参与分割,利用小班数据的属性信息确定林地和非林地区域,在林地区域再一次进行多尺度分割,并通过ESP工具确定最佳分割尺度,通过特征表达提取对象的光谱、纹理、形状、冠层高度模型(CHM)等特征信息,通过最小冗余最大相关性(mRMR)特征选择算法提取最优特征子集,且CHM在最优特征子集中。利用随机森林(RF)分类器进行年度森林采伐迹地分类提取。年度采伐迹地提取精度达到了87%,与没有CHM特征参与分类的情况对比,提取精度提高了13%。更多还原

【Abstract】 This paper using GF-1 Images to Combine Airborne LIDAR Data for Object-Oriented Classification to Extract Annual deforestation. In the object-oriented remote sensing software Ecognition. First use of forest data to participate in segmentation,Identify Forest land and non-forest areas using attribute information of small class data,Multi-scale segmentation in the forest area and determination of the optimal scale by ESP tool,Through the feature of the object extraction of the spectrum、texture、shape、CHM and their feature information,And the optimal feature subset is extracted by mutual information minimum redundancy and maximum relevance( mRMR) feature selection algorithm,CHM is in the optimal feature subset. Using Random forest( RF) classifier to extract the annual deforestation. Accuracy of extraction of annual harvest sites reached 87%. The classification accuracy is improved by 6% by comparison with the absence of CHM feature.更多还原