Swarm target networks play an important role in modern battlefields. Encrypted countermeasure technology makes it difficult to collect enemy information. Resource data acquisition is limited in non-cooperative situations that cannot participate in cluster target communication. The existing topology inference work scenarios are very different, and there is a lack of cluster network communication data sets. Therefore, this paper uses the network simulation software NS-3 to construct a more general communication network data set, realize the analysis and comparison of the same scene based on the model-driven algorithm. Then picture the communication data, and using the convolutional neural network, in the case of limited training data, the data-driven low-resource target on-off reasoning is realized. Experimental results on this dataset show that data-driven methods are more accurate for Ad hoc networks, while star networks are more suitable for model-driven algorithms.