Open AccessMulti‐label hybrid radar signals recognition and visualization based on interpretable convolutional classification layer and discriminative region suppression
Author(s) -
Si Weijian,
Luo Jiaji,
Deng Zhian
Publication year - 2022
Publication title -
electronics letters
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.375
H-Index - 146
eISSN - 1350-911X
pISSN - 0013-5194
DOI - 10.1049/ell2.12507
Subject(s) - discriminative model , interpretability , artificial intelligence , pattern recognition (psychology) , radar , computer science , feature (linguistics) , convolutional neural network , feature extraction , radar imaging , telecommunications , philosophy , linguistics
As the electromagnetic environment becomes increasingly complex, the radar receivers may receive multiple radar signals simultaneously. However, the current radar signal recognition algorithm based on deep learning can only predict a single class. The interpretable multi‐label hybrid radar signal recognition framework based on the interpretable convolutional classification layer and discriminative region suppression is proposed. The proposed interpretable convolutional classification layer is a weakly supervised localization method that can directly localize the discriminative regions of the corresponding class from the feature map of the model without additional computation. The proposed method can recognize, localize and separate hybrid radar signals in time‐frequency images, which improves the interpretability and transparency of the model. In addition, the feature pyramid network is adopted to improve the spatial resolution of the interpretable feature maps. Experiments with eight different modulation types of mixed radar signals show that the recognition accuracy for hybrid radar signals achieves 97.4% at 0 dB.