基于ResNet的智能恒虚警目标检测方法  

Intelligent Constant False Alarm Ratio Target Detection Method Based on ResNet

作　　者：张晨 叶舟 吕宇宙 方明 高永婵 ZHANG Chen;YE Zhou;LYU Yuzhou;FANG Ming;GAO Yongchan(Electronic Engineering Department,Xi’Dian University,Xi’an 710068,Shaanxi,China;Shanghai Aerospace Electronic Technology Institute,Shanghai 201109,China)

机构地区：[1]西安电子科技大学电子工程学院,陕西西安710068 [2]上海航天电子通讯设备研究所,上海201109

出　　处：《上海航天（中英文）》2022年第5期71-78,共8页Aerospace Shanghai（Chinese&English）

基　　金：航天科技基金(SAST2018-098);航空基金(20180181001);基础研究计划领域基金(2019-JCJQ-JJ-060);中国博士后科学基金(2019M653561、2020T130493);陕西省青年人才托举计划(20190104)。

摘　　要：传统恒虚警检测方法通常假设待测单元的杂波与训练杂波满足独立同分布,然而在实际雷达工作环境中,强杂波呈现出非均匀特性,沿距离单元的杂波统计特性差异变化较大,使得传统恒虚警目标检测方法在复杂环境下的检测性能下降。为解决该问题,本文从杂波统计特征提取分类出发,通过深度学习方法对杂波进行分类,提出了一种基于残差神经网络(ResNet)的智能恒虚警目标检测方法。首先,根据雷达实测杂波数据建立ResNet的训练集和测试集;其次,构建ResNet对数据进行智能特征提取,得到杂波的统计特征,并用训练好的网络对测试集进行测试;最后,以阈值可设的Softmax分类器对所得统计特征进行分类,根据分类结果实现智能恒虚警目标检测。实验结果表明:相比传统恒虚警检测方法,本文提出的方法具有自适应能力更强、检测性能更好等优点。Conventional constant false alarm ratio(CFAR)detection method usually assumes that the clutter and training clutter of the unit to be tested are independent and distributed identically. However,in the actual radar working environment,the strong clutter shows non-uniform characteristics,and the statistical characteristics of clutter vary greatly along the range cell,which makes the detection performance of the conventional CFAR detection method decline in the complex environment. In order to solve this problem,this paper proposes an intelligent CFAR target detection method based on residual neural network(ResNet) through clutter statistical feature extraction and classification by the deep learning method. First,the training set and test set of ResNet are established based on radar clutter data. Then, the ResNet is constructed to extract intelligent features from the data, and the statistical characteristics of the clutter are obtained. The trained network is used to test the test set,and the statistical features are classified by the Softmax classifier with an adjustable threshold. Finally,according to the classification results,the intelligent CFAR target detection is realized based on the classification results. Experimental results show that compared with the conventional CFAR detection method,the proposed method has better adaptive ability and detection performance.

关 键 词：雷达目标检测 残差神经网络 恒虚警检测 非均匀杂波 

分 类 号：TN957.51[电子电信—信号与信息处理]