Novel Use of PRF Sound for Radar Emitter Recognition: A Transfer Learning-Infused DCNN Study

In contemporary electronic warfare, the accurate and prompt identification of radar emitters is crucial, especially for the efficiency of electronic countermeasures. This study presents a new method that utilizes pulse repetition frequency (PRF) sound to identify radar emissions in response to the growing intricacy of modern radar systems. This study employs six transfer learning-based deep convolutional neural networks (DCNNs) to extract features. It provides a comprehensive examination of classification performance and computational efficiency across several DCNN designs. The VGG۱۶ and ResNet۵۰V۲ models achieved recognition accuracies of ۹۵.۳۸% and ۹۶.۹۲%, respectively, with training times of ۸.۰۱ seconds and ۲۱.۲۵ seconds. This study also examines the trade-offs between accuracy and computational requirements, offering a strategic understanding of the subtle dynamics of radar emitter recognition. In situations when computational complexity is not the primary concern, ResNet۵۰V۲ is the most suitable choice. Alternatively, VGG۱۶ is recommended due to its ability to compromise high accuracy and lower computing demands. This study establishes a standard for future research endeavors, which encompass enhancing the capabilities of models at a larger scale, optimizing current architectures without sacrificing accuracy, and progressing towards models that can autonomously adapt to hardware limitations. The results provide a thorough manual for choosing DCNN models that can effectively detect six different input types in various computational settings. This paves the way for creating advanced models that strike a harmonic balance between efficiency and accuracy.

Keywords:

PRF sound , radar emitter , deep convolutional neural network , extreme learning machines , gray wolf optimizer