后验概率支持向量机模型在目标分类中的应用

Posteriori probability support vector machines model with applications in target classification

针对不确定分类问题中所需的后验概率不能由传统支持向量机提供的情况,基于相对交叉熵,提出一种后验概率支持向量机建模的新方法。首先给出分类问题中交叉熵与相对交叉熵的确定方法;然后利用最小化相对交叉熵的方法建立后验概率支持向量机模型,给出了具有逆向线性搜索特点的牛顿迭代方法求解后验概率支持向量机相关模型参数的算法及思路;并通过对多种数据样本的实验,验证了后验概率支持向量机用于两类分类时的合理性与有效性。在此基础上设计了基于后验概率SVM的多类分类器,并应用于空中目标分类,实验结果表明,后验概率支持向量机可以有效提高分类正确率。

For solving the problems of that standard support vector machines do not provide posteriori probability needed in many uncertain classifications, a modeling method of probability support vector machines based on cross entropy is proposed. Firstly it gives the method of determining cross entropy and relative cross entropy in classification problems, then the probability model of support vector machines is built by minimum relative cross entropy, and the method of determining model parameters is given in detail. Some experiment results show that the posteriori probability support vector machines model is reasonable and effective. Finally, the multi-class probability support vector machines model is built, and it is used in target classification, experiments show that the posteriori probability support vector machines model can improve the correct rate of classification.