Active Radar Jamming Recognition Method Based on Dynamic Window-Length STFT Optimization and Dual-Attention Network

Jamming recognition is a prerequisite for radar anti- jamming. In complex electromagnetic environments, traditional radar jamming recognition methods struggle to effectively capture the time-frequency evolution characteristics of non-stationary signals. Aiming at the research and application of intelligent radar jamming recognition technology in complex electromagnetic countermeasure environments, and addressing the issues of poor flexibility, insufficient time-frequency aggregation of traditional fixed STFT, which makes it difficult to adapt to the dynamic characteristics of different jamming signals, as well as the heavy computational burden of traditional CNN networks that fails to meet real-time requirements, this paper uses dynamic window- length STFT to construct a high-quality time-frequency image database of nine common active jammings for fuzes. A dual- attention feature fusion mechanism-lightweight CNN classification and recognition method is proposed, which uses spatial attention to capture the time-varying trajectory of jamming pulses and combines frequency-domain attention to filter energy-dominant frequency bands, realizing adaptive fusion of multi-dimensional jamming features and achieving real-time and high-precision recognition. Experimental results show that on the dataset generated in this paper, the average recognition accuracy of the nine active jammings exceeds 98 %, and the inference speed per sample is <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\text{1. 3 9 ~ m s}$</tex>. This provides a new technical approach for the design of lightweight jamming recognition systems in dynamic electromagnetic countermeasure.