Radar Emitter Identification Using Radar Signal Sound and Deep Convolutional Neural Network Evolved by Extreme Learning Machine and Grey Wolf Optimizer Algorithm

ABSTRACT

In today's electronic warfare landscape, rapid and accurate identification of radar emitters (RE) is crucial for electronic attack missions. Advancements in radar systems, particularly phased array radars, demand faster and more efficient recognition methods. This paper introduces a sound‐based representation of radar pulse trains for RE identification, focussing on real‐time performance. Our three‐tiered methodology first employs a transfer learning‐based deep convolutional neural network (DCNN) as a feature extractor. Second, an extreme learning machine (ELM) provides fast pattern identification. Third, a grey wolf optimiser (GWO) optimises the connection weights and biases of the ELM to enhance stability. Using an authentic dataset of 10 common RE types, we evaluate two pre‐trained DCNN models (VGG16 and ResNet50V2). Experimental results demonstrate that the optimised VGG16 and ResNet50V2 models achieve recognition accuracies of 98.46% and 98.96%, respectively, with training times of 18.41 and 69.40 seconds. Notably, the VGG16‐ELM‐GWO model reduces training time by 98% compared to fine‐tuning the full DCNN, while maintaining competitive accuracy. This work demonstrates that combining ELM with nature‐inspired optimization offers a promising solution for real‐time radar emitter recognition.