雷达近场成像中SVM的目标识别方法

雷达近场成像中,在精确定位的基础上,为解决目标形状识别问题,提出了利用支持向量机（SVM）预测目标信息的方法.根据时域算法——后向投影（BP）算法和频域算法——频率波数域（F-K）偏移算法得到的场强值作为SVM的特征数据,并利用时域有限差分法（FDTD）进行仿真.仿真结果表明,基于BP算法的SVM识别方法具有特征数据提取时间长、SVM预测时间短、多目标时目标信息全和虚警较多等特征,基于F-K算法的SVM识别方法具有特征数据提取时间非常短、SVM预测时间非常短、多目标时目标漏检的特征;两者都能较好地识别目标的形状,且前者的识别能力高于后者,而后者更适合实时成像.

Target-recognition Method for Support Vector Machine on Near-field Radar Imaging

In order to solve the problems of recognizing the shapes of the targets under the premise of the accurate positioning for the near-field radar imaging, an improved method for predicting the information of the targets by the support vector machine （SVM） is proposed. The method uses the field strength obtained by the time-domain algorithm（ back-projection and BP algorithms） or the frequency-domain algorithm（ the frequency-wavenumber migration and F-K algorithms） as the feature data of the SVM, and the near-field radar imaging model is simulated by the finite difference time domain （FDTD） method. Simulation results demonstrate that the recognition method for the SVM based on the BP algorithm has some characteristics, for example, the extracting time of the feature data is long, the predicting time of the SVM is short, and the information of the targets is rich and the false alarm rate is high when there are multiple targets. While the recognition method for the SVM based on the F-K algorithm has different characteristics ,for example, the extracting time of the feature data is very short, the predicting time of the SVM is very short too, and some targets may be undetected when there are multiple targets. Both the recognition methods can recognize the shape of the targets, and the former has better identification ability than the latter, and the latter is more suitable for the real-time imaging.