基于CART算法的空间目标本体反演方法

Spatial Object Ontology Inversion Method Based on CART Algorithm

针对预警中空间群目标反演精度不高、鲁棒性较差的问题,提出了一种基于分类回归树(classification and regression tree,CART)算法的空间目标本体反演方法。首先构建分类与回归决策树,并将空间目标的入射角、光谱辐射亮度、温度等特征数据作为决策树的输入;再基于基尼系数评估数据纯度,对目标在不同温度、不同入射角下的光谱辐射亮度进行分割;最后通过在每个节点处对数据集进行划分,实现对空间目标本体的有效反演。数值对比实验表明,3种典型空间目标的决策树模型反演结果与实际情况一致,验证了所提方法的有效性。

Aiming at the problems of low precision and poor robustness of spatial group object inversion in early warning process,a spatial object ontology inversion method was proposed based on the classification and regression tree(CART)algorithm.Firstly,the classification and regression decision tree was built,the feature data such as incidence angles,spectral radiance,temperature of the spatial object were extracted as the input of the decision tree.Then the data purity was evaluated according to Gini coefficient,the spectral radiance of the object at different temperatures and incidence angles was divided.Finally,the effective inversion of the spatial object ontology was realized by dividing the data set at each node.Numerical experiments results show that the prediction results of the decision tree model for three typical spatial objects are consistent with the truth,which verifies the effectiveness of the proposed method.