基于卷积神经网络的低分辨雷达目标一步识别技术

现有低分辨雷达目标识别通常采用先特征提取、再进行目标分类的两步识别算法，存在识别率难以提高和方法泛化性不足的问题，因此提出了一种基于卷积神经网络(CNN)的低分辨雷达目标一步识别算法。该算法直接将采样数据作为输入，利用设计的一维CNN，通过卷积池化等操作自动获取数据深层本质特征，无需特征提取，实现对目标的一步识别。仿真实验结果表明:基于CNN的低分辨雷达目标一步识别方法的识别率较传统基于提取特征的两步识别方法提高了10.31%，识别时间较传统两步识别方法减少了0.142 s，充分证明了一步识别方法的有效性，为低分辨雷达目标识别问题提供了新的解决途径。

Low Resolution Radar Target One Step Recognition Technology Based on Convolutional Neural Network

In view of the problem of the existing methods of low resolution radar target recognition, usually the two step recognition algorithm is adopted to make feature extraction and target classification, which is not conducive to the improvement of recognition accuracy and the generalization of recognition methods, for the reason mentioned above, an one step recognition algorithm based on Convolution Neural Network (CNN) for low resolution radar target is proposed. This algorithm takes the sampled data as input directly, and uses the designed one dimensional CNN to automatically obtain the deep essential features of the data without feature extraction through convolution pooling and other operations, realizing the one step recognition of the target. The simulation results show that the recognition rate of the one step recognition method is 10.31% higher than that of traditional two step recognition method based on artificial feature extraction, and the recognition time of one step recognition method is 0.142s less than that of two step recognition method, which proves the effectiveness of one step recognition method. The one step recognition method provides a new solution for radar target recognition.