基于Scat-LSTM的小样本HRRP识别方法

Method for small sample HRRP recognition based on Scat-LSTM

摘要:

为提高小样本情况下的高分辨距离像(HRRP)目标识别精度,提出了一种基于小波散射变换的HRRP目标识别算法Scat-LSTM。首先,对原始信号进行小波散射变换,得到小波散射系数矩阵;然后,将该特征矩阵输入深度神经网络中进行训练和识别。实验结果表明, 在样本量充足的情况下,相比于直接使用原始信号作为输入的方法,Scat-LSTM平均识别率提升了4%,并且在训练样本量极少的情况下,也能取得比其他算法更好的识别率。

Abstract:

To improve the accuracy of high resolution range profile (HRRP) target recogni tion under small sample conditions, a HRRP target recognition algorithm Scat-LSTM based on wavelet scattering transformation was proposed. Firstly, the original signal was subjected to wavelet scattering transformation to obtain the wavelet scattering coefficient matrix. Then, this feature matrix was input into a deep neural network for training and recognition. The experimental results indicate that, given a sufficient sample size, the average recognition rate of Scat-LSTM has improved by 4% compared to methods that directly use raw signals as input. Moreover, even with extremely limited training samples, it can achieve better recogni tion rates than other algorithms.