类别不平衡条件下的雷达信号识别

在实际电子侦察过程中，由于各种原因，侦收到的不同类型信号数量相差很大，类别之间严重不平衡，常规方法在这种数据集下训练得到的分类器不能有效识别少数类。针对这一问题，首先采用栈式自编码器对中频数据进行降维和特征提取；然后在降维后的特征空间内通过多种过采样方法生成新的少数类样本，使数据集重新平衡，并利用再平衡后的数据集训练支持向量机分类器；最后采用F分数和受试者工作特征（Receiver Operating Characteristic,ROC）曲线两种评价方法对分类效果进行评价。实验结果表明，通过过采样处理，分类器对少数类的识别性能有所提升。

Radar signal recognition under category imbalance condition

In the actual electronic reconnaissance process,due to various reasons,the number of received signals of different types varies greatly,and there is a serious imbalance between classes.To solve this problem,this paper uses stacked autoencoder to reduce the dimension and extract the features from the intermediate frequency data,and then generates a few new samples in the reduced dimension feature space by various over-sampling methods,so that the data set is rebalanced.Then the support vector machine classifier is trained by the rebalanced data set.Finally,the classification effect is evaluated by F-score and receiver operating characteristic(ROC) curve.The experimental results show that the performance of the classifier for a few classes is improved through oversampling.