在雷达HRRP识别中多特征融合多类分类器设计

在传统线性关联向量机的基础上,设计了一种多特征融合的多类分类器.该分类器基于多类Probit回归模型将传统的两类线性关联向量机推广为多类关联向量机,利用线性关联向量机的特征选择功能,对融合的高维特征向量进行降维和合理的幂次扩展,使线性关联向量机具有构造非线性分类界面的能力,以保证对非线性多类分类问题稳健的融合识别性能.针对雷达高分辨距离像目标识别问题,提取3种平移不变特征,使用提出的多特征融合的多类分类器在基于实测数据的识别实验中得到了稳健的融合识别结果.

Multi-class classifier design for feature fusion in radar HRRP recognition

A multi-class classifier for feature fusion is designed based on the traditional linear Relevance Vector Machine (RVM). The proposed classifier extends the binary RVM to multi-class RVM based on the multinomial Probit regression model, utilizes the feature selection property of the linear RVM to reduce the dimensionality of the fused feature vector, and makes the linear RVM framework have the ability to form the nonlinear classification boundary via the rational power extension. All of these properties can ensure the robust fusion recognition performance in the case of nonlinear multi-class classification. For the application of radar target recognition using the high-resolution range profile (HRRP), the experimental results on the measured data show that the proposed multi-class classifier with the three translation-invariant features extracted from HRRP data can achieve robust recognition performance.