Abstract: A multi-level multifunctional radar behavior level representation model is established to address the problems of complex and varied styles, incomplete overall feature representation, and insufficient ability to provide key information in signal level analysis of multifunctional radar. A fusion network structure based on parallel processing of one-dimensional deep convolution neural network and gated recurrent unit is proposed. On the basis of using multi-level models to clearly and effectively characterize and analyze the behavior of multifunctional radar, combined with the advantages of two networks in local depth feature extraction and global time-series feature extraction, the behavior identification of typical functions of multifunctional radar has been achieved. The simulation experiment results show that, with a high degree of parameter interleaving, the network achieves a behavior recognition accuracy of 95. 6% for the four typical functions of multifunctional radar, which proves that the proposed parallel network algorithm has good application prospects in the field of reconnaissance intelligence analysis.