Identification of Radar Emitters Using Radar Signal Sound and Transfer Learning Deep Convolutional Neural Networks

Identifying threat radars is one of the critical tasks of electronic intelligence and support systems and a fundamental challenge in modern electronic warfare. With the rapid advancement of radar technologies and the emergence of advanced multi‐function radars such as phased array radars, accurate and fast identification of these radars has become increasingly important. This importance stems from the decisive role of radar information in military decision‐making, situational awareness creation and the implementation of countermeasures. To address this need, this study presents an innovative approach that uses radar signal sound for radar identification for the first time. In this study, six pre‐trained deep convolutional neural networks (DCNNs) were used, which were trained and evaluated using a real dataset. For stable and reliable evaluation, a 5‐fold cross‐validation was used. The identification results show that the ResNet50V2 and VGG16 models have the best identification performance with accuracies of 97.25% ± 1.15% and 96.12% ± 1.28%, respectively. The VGG16 model shows superior computational efficiency (25.8 million parameters, 11.80 s per epoch), whereas the ResNet50V2 model provides the best accuracy with sufficient resources. Comparison with baseline methods (support vector machine (SVM) with an accuracy of 89.29% and multi‐layer perceptron (MLP) from scratch with an accuracy of 61.29%) and an advanced meta‐learning method (88.60%) confirms the superiority of the proposed approach.