系统工程与电子技术

• 传感器与信号处理 • 上一篇    下一篇

基于改进半监督朴素贝叶斯的LPI雷达信号识别

符颖, 王星, 周一鹏, 范翔宇   

  1. 空军工程大学航空航天工程学院, 陕西 西安 710038
  • 出版日期:2017-10-25 发布日期:2010-01-03

Recognition of LPI radar signals based on revised semi-supervised Naive Bayes

FU Ying, WANG Xing, ZHOU Yipeng, FAN Xiangyu   

  1. Aeronautics and Astronautics Engineering College,Air Force Engineering University, Xi’an 710038, China
  • Online:2017-10-25 Published:2010-01-03

摘要:

针对先验信息残缺的非合作电子对抗背景下的低截获概率雷达信号识别问题,提出一种基于改进的半监督朴素贝叶斯的识别算法。该算法首先提取出4种低截获概率(low probability of intercept,LPI)雷达信号的双谱对角切片作为识别特征;针对传统的半监督朴素贝叶斯(semi-supervised Naive Bayes,SNB)在更新训练样本集过程中会产生迭代错误的不足,利用改进的SNB(Revised SNB, RSNB)算法构建分类器,完成对测试样本的识别。该方法通过在无标记样本集生成的置信度列表中选取置信度较高的样本添加到有标记样本集中,再利用预测后的分类结果对分类器参数(即特征期望向量m i和方差向量σ i)进行改进,有效解决了传统算法分类精度低且分类性能不稳定等缺点。理论分析和仿真结果表明,在LPI雷达信号识别问题,相比于SNB算法和传统的主成分分析加支持向量机法(principal component analysis-support vector machine, PCA-SVM),该算法具有更高的分类识别率和更好的分类性能。

Abstract:

In order to solve incomplete prior information of low probability of intercept (LPI) radar in non-cooperative electronic countermeasure environment, a novel recognition algorithm based on revised semi-supervised Naive Bayes (RSNB) is proposed. The RSNB algorithm extracts bispectrum diagonal slices of four LPI radar signals as the recognition feature. To overcome disadvantages of traditional semi-supervised Naive Bayes which comes from repeated errors in updating sample sets, it uses revised semi-supervised Naive Bayes to construct the classifier, and then completes the recognition of tested sample sets. RSNB selects those samples with high degree of confidence which comes from generated confidence list in unlabeled samples sets so as to add them to labeled samples sets, then improves the classifier parameters by using predicted results. It can work out low recognition rate and unstable classification performance effectively by using the revised semi-supervised Naive Bayes. The simulated results indicate that, the RSNB has higher recognition rate and better classification performance when compared with traditional SNB algorithms and the principal component analysis-support vector machine algorithm in LPI radar recognition.