Frequency Hopping Signal Parameter Estimation Algorithm Based on Time-Frequency Features and Semantic Segmentation

Under non-cooperative conditions in wireless real-time communications toward 6G, frequency hopping (FH) signal reconnaissance remains a challenging task, particularly in complex electronic environments characterized by multisource interference, signal overlap, and low signal-to-noise ratio (SNR). To address FH signal parameter estimation in such scenarios, this paper proposes a deep learning-based extraction and estimation method, deepLabv3+ network is introduced for semantic segmentation of FH signals. The approach employs Gaussian filtering for time-frequency diagram denoising and utilizes a dual-branch DeepLabv3+ network to automatically extract FH signal segments through robust feature extraction. By leveraging the distribution of segmented hops in the time-frequency domain, key parameters such as hop period, transition timing, and frequency set are accurately estimated. Experimental results demonstrate that the proposed method achieves high estimation precision while significantly enhancing robustness in low-SNR environments.