基于卷积原型网络的通信辐射源个体开集识别方法  

Open-Set Emitter Identification Based on Convolutional Prototype Network

作　　者：刘人玮 李天昀[1] 章昕亮 沈小龙 龚佩 LIU Renwei;LI Tianyun;ZHANG Xinliang;SHEN Xiaolong;GONG Pei(Information Engineering University,Zhengzhou 450001,China)

机构地区：[1]信息工程大学,河南郑州450001

出　　处：《信息工程大学学报》2023年第2期140-149,共10页Journal of Information Engineering University

摘　　要：针对通信辐射源个体开集识别问题,提出一种基于卷积原型网络的辐射源个体开集识别方法。将接收到的信号进行信噪比估计并处理为灰度矢量图输入至卷积原型网络计算出特征点。分析不同损失函数和判决准则,选取三维原型的特征提取层、DCE作为损失函数、DR作为判决准则的开集识别模型。对该方法进行了仿真实验,结果表明,该方法在开集识别场景下比传统CNN预测概率更有优势,归一化准确率提升约20%,证明该方法的有效性。An open-set identification method based on convolution prototype network is proposed to solve the problem of communication emitter identification.The received signal is estimated by signal-to-noise ratio and processed as a gray signal trajectory image,which is input into the convolutional prototype network to calculate the feature points.Feature extraction layer of 3D prototype,DCE as loss function and DR as judgment criterion is selected to open-set recognition model by analyzing dif-ferent loss functions and judgment criteria.The simulation results show that this method has more advantages than traditional CNN prediction probability in open-set recognition,and the normalized accuracy is improved by about 20%,which proves the effectiveness of this method.

关 键 词：开集识别 辐射源识别 矢量图 卷积原型网络 

分 类 号：TN911.7[电子电信—通信与信息系统]