Deep learning-based radar active jamming open set recognition and unknown jamming clustering methods

To address the problem of radar's recognition of unknown types of active jamming in complex electromagnetic environments, a deep learning-based radar active jamming open-set recognition and unknown jamming clustering method is proposed. First, a radar active jamming recognition network based on multi-layer channel attention mechanism is designed by introducing residual module, Inception module, and attention mechanism module; second, the time-frequency map and Range-Doppler map of jamming signals are used to form two input branches, and the relative entropy is obtained according to the respective recognition probability distributions as the confidence of the recognition results, and through the recognition of the probability distribution of the maximum index and the relative entropy The threshold is set by the voting of the maximum index of the recognition probability distribution and the relative entropy, which realizes the open-set recognition of the unknown type of jamming; finally, the data adaptive spatial clustering algorithm is designed by analyzing the feature principal components obtained from the mapping of the deep learning network, downsizing and extracting the feature parameters that account for more than 95% of the total number of features, and realizing the clustering of the unknown type of jamming. The simulation data classify 14 types of jamming signals into 8 types of known jamming and 6 types of unknown jamming, and can realize more than 91.4% of active jamming open set recognition and effective clustering of unknown jamming under the condition that the jamming noise ratio is more than 5 dB.