在电子战、无线网络安全等军事和民用领域中,特定辐射源识别(SEI)具有极高的应用价值。传统方法以人工提取特征为主，依赖先验知识泛化性差；深度学习方法多以含有二维信息的图像为输入，易遗漏关键信息。针对以上问题，提出了以数字频谱余晖图作为深度学习模型输入的解决手段，从而实现了SEI 任务。首先,搭建辐射源信号探测及数据采集系统，获取Wi-Fi 辐射源信号的数字频谱余晖图，建立首个基于数字频谱余晖图的SEI 数据集；其次，利用该图包含信息更丰富的特点,将信号识别问题转化为目标检测问题；最后，在Wi-Fi 辐射源识别数据集(WFEID)上进行了实验验证。实验结果表明，YOLOv5s 在WFEID 上P、R、F1 和mAP等指标均能达到87. 5%以上，从而证明了以数字频谱余晖图作为深度学习模型的输入在SEI 任务中是有效的。

In military and civilian fields such as electronic warfare and wireless network security, specific emitter identification (SEI) has extremely high application value. The traditional methods are mainly based on manual feature extraction, which rely on prior knowledge and have poor generalization. Deep learning methods mostly use images containing two-dimensional information as input, which is easy to miss key information. In order to solve the above problems, a solution method using the afterglow map of the digital spectrum as the input of the deep learning model is proposed, so as to realize the SEI task. Firstly, an emitter signal detection and data acquisition system are built to obtain the afterglow map of the digital spectrum for the Wi-Fi emitter signal, and the first SEI dataset based on the afterglow map of the digital spectrum is established. Secondly, the problem of signal recognition is transformed into the problem of target detection by using the feature that the image contains more information. Finally, experimental verification is performed on the Wi-Fi emitter identification dataset (WFEID). Experimental results show that the P, R, F1 and mAP of YOLOv5s can reach more than 87. 5% on WFEID, which proves the effectiveness of the method using the afterglow map of the digital spectrum as the input of deep learning model in tasks of specific emitter identification.

TN971;TP391.4