采用最优集成学习的小样本电磁脉冲信号分类

Small samples electromagnetic pulse classification using an optimal ensemble learning method

闪电与核爆电磁脉冲分类是核检测系统中的关键问题,其主要难点便是其正负样本不均衡程度可达到104,因此我们提出了一种名为SMALLBAG的集成学习分类方法。针对小样本问题,通过对少数类样本进行数据增强和多数类样本重采样的方法重新构建新的训练数据集,分别提取时域、频域、小波域的特征以表征信号。针对样本不均衡问题,提出了基于新采样数据集的集成学习方案,减少样本不均衡影响同时提高分类准确率。该模型能够在保证准确率的同时保证实时性要求,试验结果显示识别准确率可达99.99%,测试速度为每个样本0.67 ms。

The classification of nuclear and lightning electromagnetic pulse is a key problem in nuclear identification system,in which the main challenge is that the level of class imbalance can be as huge as,so SMALLBAG,a novel ensemble learning method,was proposed.Firstly,a preprocessing procedure was developed to rebuild training set,in which data augmentation methods were used to generate more minority class samples and resampling scheme was proposed to approach class balance.Secondly,feature extraction was performed in time,frequency and wavelet domains,which were used to characterize the signal.Finally,the ensemble learning method was proposed to alleviate the influence of class imbalance and improve the performance of identification.Experimental results show that the simplicity of the proposed learning method ensures the identification accuracy and real-time requirements at the same time,i.e.99.99%identification accuracy and 0.67 ms testing time per sample.