雷达海上目标双通道卷积神经网络特征融合智能检测方法

受复杂海洋环境影响,基于统计理论的海面目标检测方法由于假设条件不成立,在实际应用中难以实现高性能检测,本文从特征提取分类角度,通过深度学习分类方法对目标和杂波的雷达回波信号进行二元分类,提出了一种基于双通道卷积神经网络(DCCNN)的雷达海上目标智能检测方法。首先,对实测海杂波和目标雷达信号进行预处理,得到信号的时间-多普勒谱和幅度信息;然后,构建DCCNN对预处理得到的数据进行智能特征提取,得到信号的特征向量,并对不同特征提取模型性能进行测试;最后,通过阈值可设的Softmax分类器作为检测器对特征向量进行分类,实现虚警率的控制。测试结果表明:与传统的单通道CNN以及无虚警控制Hog-SVM分类算法相比,基于二维卷积核VGG16和一维卷积核LeNet的DCCNN特征提取模型和softmax分类器可实现更高的检测性能,并可以实现虚警率控制,为复杂海杂波背景下目标智能检测提供了新的技术途径。

Dual-channel Convolutional Neural Networks Feature Fusion Method for Radar Maritime Target Intelligent Detection

based on statistical theory is difficult to achieve high-performance detection in practical application, since the assumptions of clutter statistical distributions are not met. In this paper, from the perspective of feature extraction and classification, the radar echoes of the target and clutter are classified with deep learning classification method. An intelligent detection method for radar maritime targets based on dual channel convolutional neural network (DCCNN) is proposed. Firstly, the measured sea clutter and the target radar signal are preprocessed to obtain the time-Doppler spectrum and amplitude information of the signal. The DCCNN is constructed to extract the intelligent features of the preprocessed data, obtain the eigenvectors of the signals, and test the performance of different feature extraction models. Finally, the Softmax classifier, whose threshold can be adjusted, is used as a detector to classify the feature vectors to realize the control of the false alarm rate. The test results show that, compared with traditional single-channel CNN and Hog SVM classification algorithm, DCCNN feature extraction model based on two-dimensional convolution kernel VGG16 and one-dimensional convolution kernel LeNet and softmax classifier can achieve higher performance and can achieve the control of false alarm. It provides a new technical approach for the target intelligent detection in the complex sea clutter background.