基于改进VGG16的超短波时频谱图分类方法

现代战场电磁环境日益复杂，战术通信网台主要集中在超短波频段。而未来的技术侦察对智能化、大数据处理的支撑需求越来越强烈。为实现超短波盲信号的分类，提出了一种将盲信号的时频谱图与优化后VGG16网络相结合的分类方法。该方法首先将电磁战场中实际采集到的超短波盲信号转换为时频谱图，然后通过迁移学习将其与优化后的VGG16卷积神经网络结合起来，并将空洞卷积引入网络，完成了对超短波盲信号的分类。实验结果表明，优化后的VGG16网络比原网络有更高的准确率，达到了93.1%。当将空洞卷积引入到优化后的VGG16网络的第7层和第10层时，识别率达到最高为92.2%，学习时间减少了34.1%，大大减少了模型的训练时长，验证了空洞卷积在超短波盲信号分类识别上的有效性。

Classification method of ultrashortwave time spectrum based on improved VGG16

The electromagnetic environment of modern battlefield is becoming more and more complex, and the tactical communication network stations mainly focus on the ultra-short wave frequency band. The future technical reconnaissance has an increasingly strong support demand for intelligent and big data processing. In order to realize the classification of ultra-short wave blind signals, a classification method combining the time spectrum of blind signals with the optimized VGG16 network is proposed. The method first converts the actual VHF blind signals collected in the electromagnetic battlefield into time-spectrum maps, then combines them with the optimized VGG16 convolutional neural network through transfer learning, and introduces the cavity convolution into the network to complete the classification of VHF blind signals. Experimental results show that the optimized VGG16 network has a higher accuracy than the original network, reaching 93.1%. When the cavity convolution is introduced into the 7th and 10th layers of the optimized VGG16 network, the recognition rate reaches the highest of 92.2%, and the learning time is reduced by 34.1%, which greatly reduces the training time of the model, and verifies the effectiveness of cavity convolution in VGG16 blind signal classification and recognition.