Recognition of Communication Relationship Based on the Spectrum Monitoring Data by Improved VGGNET

Abstract

The communication relationship can reflect the hidden information of the communication network, which is of great significance for discovering important nodes in the network. To overcome the difficulty of manually extracting expert features, this paper uses deep learning methods to study the communication relationship recognition. First, use the deep learning model to classify the spectrum data directly, and model the communication relationship as a classification problem with time feature data for processing. It is found that the neural network model is easy to fall into a local minimum; in order to limit the impact of the local minimum problem on recognition In this paper, combining the rules of frequency hopping communication to process the data, make the neural network take as few tasks as possible, and then propose the second design scheme, the communication time series classification scheme, and the final recognition rate reaches 97% on the test set. This article uses long and short memory networks and convolutional neural networks to conduct experiments. Among them, the improved VGG network structure has the best recognition rate in communication problems. The factors that affect the recognition rate of neural networks in the identification of communication relationships are discussed in depth, and suggestions on how to adjust these factors are given based on theory and experiment.