Prototype learning based open set recognition algorithm for interference signals

Existing deep learning-based interference signal recognition mod-els focus on closed set recognition (CSR), whose training set and test set have the same classes of data. However, it will face the problem of identifying un-known interference signals as known interference signals in real-world applica-tions. Therefore, it is necessary to ensure that the model will accurately classify the known classes and reject the unknown classes. We propose an Open Set Recognition (OSR) algorithm based on open prototype learning (OPL) and a time-frequency fusion feature extraction network (TFFNet) for interference sig-nals, namely OPL-TFFNet. Firstly, an OPL method for open set recognition is proposed. Based on the prototype learning strategy, the closed-set classifier and the open-set classifier are combined by setting the unknown class prototype. The OPL method can adaptively divide the boundaries of known interference and unknown interference. Secondly, TFFNet is designed by using the residual structure. The depth features of the interference signals in the time domain and frequency domain are extracted, respectively, and inputted to the network with masked data-augmented, and the feature fusion is performed through nonlin-ear transformation after splicing. That TFFNet effectively improves the feature extraction ability of the network under higher noise. Lastly, we carried out sim-ulation experiments on nine interference types. The simulation results show that the proposed OPL-TFFNet method has good performance both on CSR and OSR. Compared with the existing CG-Encoder algorithm, the CSR performance is improved by 3% in the case of low jamming-noise-ratio (JNR), and when the openness is 0.03, the OSR performance is improved by 3,-,4%. When the openness is 0.397, the OSR performance is improved by 7%.