基于数据增强的小样本通信干扰识别技术

Few-shot Communication Jamming Recognition Technology Based on Data Augmentation

干扰识别是通信抗干扰过程的前置环节,为抗干扰决策和波形选择提供关键的先验知识,是抗干扰成功与否的关键步骤。在复杂的电磁条件下,对干扰信号进行大量标记往往是困难的,并且对干扰识别的实时性要求很高。针对上述问题,研究了一种基于数据增强的小样本干扰信号识别技术,以解决小样本条件下分类器训练过程的欠拟合问题。小样本条件下,在正态贝叶斯分类器和朴素贝叶斯分类器训练过程中,分别设计了两种数据增强方法增加训练样本以提高分类器性能。仿真结果表明,采用数据增强的干扰识别技术能够显著提高小样本条件下干扰识别的准确率。

Jamming recognition is the pre-stage of the communication anti-jamming process,providing key prior knowledge for anti-jamming decision-making and waveform selection,and is a key step for the success of anti-jamming.under complex electromagnetic spectrum conditions,it is often difficult to mark a large number of jamming signals in a short time.Also,the real-time requirements for jamming recognition are very high.To solve these problems,this paper studies a few-shot jamming signal recognition technology based on data augmentation to solve the underfitting problem of the classifier training process.Specifically,in the training process of the normal Bayesian classifier and the naive Bayesian classifier under the few-shot condition,two data augmentation methods are designed to increase the training samples,in order to improve the performance of the classifier.Simulation results show that the use of data augmentation can significantly improve the accuracy of jamming recognition under few-shot conditions.