论文部分内容阅读
激光扫描与测距系统(LIDAR)所获取的点云数据能够表达地物的三维信息,而光谱相机能够获得同场景的四个波段的多光谱信息。二者从不同的侧面表现了地物的特征,但不同特征对分类精度的贡献具有较大的差异。提取不同类型的地物特征,将特征分成四组;以随机森林为分类框架,得到不同特征子集的重要性测度和每个像元对各类别的隶属度;提出自适应D-S证据方法对各特征子集的分类证据进行合成,实现地物类别信息提取。充分利用两分类器的优点挖掘分析遥感不确定性信息,实验结果表明,分类精度达到90%,能够达到应用要求。但通过进一步分析,由于仍然是像元级的处理,初始分类结果在特殊区域存在混淆现象,影响了分类精度,通过采用基于空间限制的方法对混淆区域分类结果进行优化,提高了分类精度。
The point cloud data acquired by Laser Scanning and Ranging System (LIDAR) can express the three-dimensional information of the object, while the spectrum camera can obtain the multi-spectral information of the four bands of the same scene. Both show the features of features from different aspects, but the contributions of different features to the classification accuracy are quite different. The features of different types of features were extracted and the features were divided into four groups. Using the random forest as the classification framework, the importance measures of different feature subsets and membership degree of each pixel to each category were obtained. Feature subset of the classification of evidence synthesis, to achieve the type of object information extraction. Making full use of the advantages of the two classifiers to mine and analyze the uncertainty of remote sensing information, the experimental results show that the classification accuracy can reach 90% and meet the application requirements. However, through further analysis, because of the still pixel-level processing, the initial classification results are confused in the special area, which affects the classification accuracy. The classification result is improved by using the space-based restriction method to improve the classification accuracy.