Time-Frequency Image-Based Overlapping LPI Radar Signal Detection and Recognition with LPI-YOLO

Area of Science:

• Electrical Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Accurate recognition of overlapping low-probability-of-intercept (LPI) radar signals under low signal-to-noise ratio (SNR) conditions is crucial for electronic reconnaissance.
• Existing time-frequency analysis methods face challenges in separating and identifying multiple overlapping signal components.

Purpose of the Study:

• To develop an enhanced deep learning model for improved recognition of LPI radar signals in low SNR environments.
• To address the limitations of current methods in handling overlapping signals and computational complexity.

Main Methods:

• An enhanced YOLOv8-based model, LPI-YOLO, was developed for processing short-time Fourier transform (STFT) spectrograms.
• Integration of a Coordinate Attention (CA) mechanism and a Transformer encoder to enhance feature representation and contextual modeling.
• Adoption of lightweight GSConv-based bottlenecks and a simplified SimSPPF module to reduce computational load.

Main Results:

• The LPI-YOLO model achieved competitive performance on a simulated dataset with six modulation types, even at SNR levels as low as -12 dB.
• Demonstrated improvements in accuracy for challenging modulation classes compared to the standard YOLOv8n model.
• The proposed model maintained lower computational complexity (parameters and FLOPs).

Conclusions:

• The LPI-YOLO model offers a promising solution for accurate LPI radar signal recognition in adverse low SNR conditions.
• The integration of attention mechanisms and efficient architectural components enhances performance and reduces computational cost in electronic reconnaissance.