Multi-dimensional Feature Fusion Modulation Classification System Based on Self-training Network

摘要

To solve the problem that the single feature extraction method cannot fully express the radar signal at low SNR and the large-scale deep learning network cannot deal with small sample size of radar signal, this paper proposes a multi-dimensional feature fusion modulation classification system, which can classify radar signals including CW, BPSK, LFM, COSTAS, FRANK, T1, T2, T3 and T4. The machine could extract time-frequency feature of radar signal automatically through small self-training network. Combined with the idea of multi-dimensional feature fusion, the time-frequency entropy feature, the higher-order statistics feature and network self-extraction feature are normalized and fused by non-negative matrix factorization (NMF), which improves the classification performance of the proposed system at low SNR. The simulation results show that the recognition rate of the proposed system is 78% at - 3 dB. Compared with the traditional method, the recognition rate of proposed system has a significant improvement.