地物分类是发挥机载LiDAR在三维智慧城市建模、特殊区域地形地貌测绘等应用中独特优势的必要前提，基于证据推理的多特征组合是实现大量LiDAR数据地物分类的有效快速化智能方法。目前问题是分类推理模型无法协调用于描述类别属性的证据间的不一致性，造成分类精度低甚至分类错误；原因是现有特征对混淆区的辨识能力有限、缺乏多个分类特征间的调和机制。为此，项目以机载LiDAR数据地物分类为对象，以可能性理论、多属性决策等智能方法为手段，研究1）高辨识特征及其可能性分布构造，包括特征的地物辨识性能、分类精度影响要素、构建复合衍生特征、特征分布构造及其特性；2）多特征的分布合成方法，包括建立可能性分布合成规则、多合成规则组合、面向分类需求的分布合成方法；3）基于分布合成推理的地物分类模型，包括分布合成方法对比和多分类特征间的性能调和；4）模型验证。通过研究为提高机载LiDAR数据地物快速分类的精度探索新途径。

The land-cover classification is a necessary precondition for the unique advantages of airborne Light Detection And Ranging (LiDAR) system in three-dimensional smart city modeling and special area topography mapping. The multi-feature combination based on evidence reasoning is an effective and rapid intelligent method for realizing the land-cover classification of a large number of LiDAR data. The current problem is that the classification reasoning model cannot coordinate the inconsistency between the evidences used to describe the category attributes, resulting in low classification accuracy or even classification errors. The reason is that current features have limited ability to identify obfuscated areas and lack coordination mechanisms among multiple classification features. To solve this problem, the project studies the following four aspects, with the project classification of airborne LiDAR data as objects and intelligent methods such as possibility theory and multi-attribute decision making as the major means. First, the high identification features and their possible distribution structures are mainly considered, including feature recognition performance, classification accuracy influence factors, construction of composite derivative features, and feature distribution structures and their characteristics. Second, the method of multi-feature distribution synthesis is explored, including the establishment of possibilities distribution-synthesis rules, multi-synthesis rule combination, and distribution synthesis method for classification requirements are studied. Third, the model of land-cover classification based on distributed synthesis reasoning, including comparison of distribution synthesis methods and performance reconciliation between multi-classification features are studied. Finally the model is verified. Through the above research, the project explores new ways to improve the accuracy of fast land-cover classification of airborne LiDAR data.