Efficient Open-Set Recognition for Interference Signals Based on Convolutional Prototype Learning

Interference classification plays an important role in anti-jamming communication. Although the existing interference signal recognition methods based on deep learning have a higher accuracy than traditional methods, these poor robustness while rejecting signals of unknown classes open-set (OSR). To ensure known and rejection rate OSR, we propose new hollow convolution prototype (HCPL) which inner-dot-based cross-entropy loss (ICE) center are used to update prototypes periphery feature space so that internal is left for class samples, radius reduce impact norm classes. Then, hybrid attention reuse net (HAFRNet) was designed, contains structure domain module (HDAM). A simple DenseNet without transition layer. An HDAM can recalibrate both time-wise channel-wise responses by constructing global matrix automatically. We also carried out simulation experiments nine types, include single-tone jamming, multitone periodic Gaussian pulse frequency hopping linear sweeping second BPSK modulation noise jamming QPSK jamming. The results show proposed method has considerable performance When JNR −10 dB, 2–7% other algorithms under different openness. openness 0.030, plateau reaches 0.9883, GCPL 0.9403 CG-Encoder 0.9869; when 0.397, more 0.89, 0.8102 0.9088. However, much worse low JNR. In addition, requires less storage resources lower computational complexity CG-Encoder.