Selecting Features Based on Information Loss for Radar Target Recognition

摘要：

As a feature of wideband radar target echoes, high resolution range profile (HRRP) always has a relatively high feature dimension along the radar line-of-sight (LOS), and its joint probability density estimation may be inaccurate when the samples are insufficient. Feature selection is an efficient way to reduce feature dimensions, and correct classification rate (CCR) is a common used objective function for feature selection. However, CCR can not afford the information loss caused by the increase of feature dimensions when the samples are finite. To compensate the shortcomings of CCR, a new objective function named CCR lower bound is introduced in HRRP feature selection. CCR lower bound is calculated by linearly combining the ordinary CCR and information loss of Mahalanobis distance. Experiment results based on a dataset of 3 aircrafts X-Band radar HRRPs show that the proposed approach works well for HRRP feature selection and target recognition.

作者: Changhong Wei

作者单位: Depart. Of Training, Automobile Management Institute, Bengbu 233011, China

会议类型: 国际会议

会议名称: 2010 4th International Conference on Intelligent Information Techonlogy Application(第四届智能信息技术应用国际学术研讨会 IITA 2010)

会议地点: 秦皇岛

会议语种:英文

页码: 77-80

在线出版日期: 2010-11-05（万方平台首次上网日期，不代表论文的发表时间）