论文部分内容阅读
针对目前支持向量机多类分类方法存在的缺点,文中对支持向量机的高斯核函数进行改进,并提出一种结合留一法和单一验证法的参数选择新方案。基于3种雷达目标的HRRP数据,设计了相应的预处理算法,利用改进的SVM分类法用于高分辨距离像的雷达目标识别。从实验目标姿态稳定性、训练集大小稳定性和抗噪能力三个方面验证改进SVM的稳健性。
Aiming at the shortcomings of the current multi-class classification methods for SVM, this paper improves the Gaussian kernel function of SVM, and proposes a new parameter selection scheme that combines the one-leave-only method and the single-validated method. Based on HRRP data of three kinds of radar targets, a corresponding preprocessing algorithm is designed and an improved SVM classification method is applied to radar target recognition of high resolution range images. The robustness of the improved SVM is verified from three aspects: attitude stability of experimental target, stability of training set size and noise immunity.