基于深度多尺度一维卷积神经网络的雷达舰船目标识别

Radar HRRP Target Recognition Based on Deep Multi-Scale 1D Convolutional Neural Network

为满足雷达舰船目标识别的高实时性和高泛化性的需求,该文提出了一种基于深度多尺度1维卷积神经网络的目标高分辨1维距离像(HRRP)识别方法。针对高分辨1维距离像特征提取难的问题,所提方法通过共享卷积核的权值,使用多尺度的卷积核提取不同精细度的特征,并构造中心损失函数来提高特征的分辨能力。实验结果表明,该模型可以显著提高目标在非理想条件下的识别正确率,克服目标姿态角敏感性问题,具有良好的鲁棒性和泛化性。

In order to meet the demand for high real-time and high generalization performance of radar recognition, a radar High Resolution Range Profile (HRRP) recognition method based on deep multi-scale one dimension convolutional neural network is proposed. The multi-scale convolutional layer that can represent the complex features of HRRP is designed based on two features of the convolution kernels which are weight sharing and extraction of different fineness features from different scales, respectively. At last, the center loss function is used to improve the separability of features. Experimental results show that the model can greatly improve the accuracy of the target recognition under non-ideal conditions and solve the problem of the target aspect sensitivity, which also has good robustness and generalization performance.