Abstract
Accurately classifying and deinterleaving overlapping radar signals presents a significant challenge in complex environments, such as electronic warfare. Traditional methods, such as spectrogram-based analysis, often struggle to differentiate radar signals with similar scan patterns, particularly under low signal-to-noise ratio (SNR) conditions. To address these limitations, we propose a novel two-stage classification framework that combines amplitude pattern (AMP) analysis and visibility graphs to enhance the accuracy and efficiency of radar signal classification. In the first stage, AMP analysis groups radar reception signals into broad categories, which reduces noise and isolates signal features. In the second stage, a visibility graph technique is applied to refine these classifications, enabling the practical separation of radar signals with overlapping or similar amplitude features. The proposed method is particularly effective in handling complex scans, such as the Palmer series, which blends search and tracking patterns. Deep learning models, including GoogLeNet and ResNet, are integrated within this framework to improve classification performance further, demonstrating robustness in low-SNR and multi-signal environments. This approach offers significant improvements over conventional methods, providing enhanced performance in differentiating radar signals across various scanning patterns in challenging multi-signal environments.