二叉树支持向量机(SVM)是一种针对多类问题的有效分类器,具有结构简单、训练快的特点,但二叉树SVM容易出现误差积累,且不能输出识别结果的置信度。文中设计了一种基于隶属度计算的二叉树SVM分类器,首先,该分类器利用方差和最小准则选择节点,将多类问题转化为偏二叉树SVM分类问题,避免了误差积累,然后,利用特征变换空间的类中心和类半径,计算出样本结果的置信度,使得二叉树SVM分类器能够输出模糊结果。将上述二叉树SVM分类器应用于弹道目标的RCS特征识别,仿真结果表明了该方法的有效性。

Binary tree SVM classification is effective on multi-class classification. It wins popularity due to simple structure and fast training, but easily appears error accumulation and cannot give the membership degree. A new Binary tree SVM classification is studied in this paper. Firstly, the new classification uses minimum standard difference to choose tree node, which can avoid error accumulation. Then, it uses class center and class radius in character transform space to compute membership degree. Finally, the fuzzy recognition result is given based on membership degree. The proposed classification is applied to ballistic target recognition of RCS data. The results verified the effectiveness of the new classification.