基于非完备贝叶斯网络的车型识别方法

针对BP神经网络等车型识别算法不能很好地适应我国车型复杂状况，分析了贝叶斯网络（IDS-BN）算法中完备数据集的问题，提出了基于非完备数据集的贝叶斯网络车型识别方法和车型识别系统结构模块，基于该模块拟定了车型特征变量．构建了车型识别网络模型，并给出了模型参数学习、车型分类器等算法，包括MDL评分和贪婪搜索的结构学习、EM参数学习、随机模拟采样推理等。实验表明，该方法识别率较高，鲁棒性好，满足我国车型识别的实际要求。

Method for Classifying Vehicles with Bayesian Network Based on Incomplete Data Sets

There are some disadvantages of classifiers like BP neural network and Bayesian network method based on complete data sets. First, this paper discusses the complete data sets. Then, it proposes a method for classifying vehicles with Bayesian network based on incomplete data sets, draws up vehicle classification variables, constructs network model of vehicle classification and gives algorithms of model parameter learning and vehicle classifiers, which include MDL scoring and greedy search, EM parameter learning, and stochastic simulation of causal models for probabilistie inference. Experiments show that it has a high accuracy and a good robustness.