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。

Point cloud data acquired from the light detection and ranging system (LIDAR) can express threedimensional information of land-cover feature, while spectral cameras can acquire multispectral information of four bands for the same scene. Both systems show land-cover features from different perspectives. Different features can contribute classification accuracy differently. Different types of land-cover features are extracted, and divide them into four groups. Importance measures of different feature groups and each pixel degree of membership to each class can be obtained based on random forest classifier.Land-cover class labels are obtained through proposing adaptive weighted D-S evidence theory to composite the classfication evidence of each single feature set. Takes full advantage of two classifiers to extract and analyze remote sensing indefinite information. The experimental results indicate that classification accuracy reaches to 90% and meets the requirement of application. But through the further analysis, because of the pixel level processing, there are much confusion in special area to decrease the classification precision in initial classification results. Optimize the classification results in confusional area by adopting space constraint ,there by classification precision is improved.