With the rapid development of wireless communication technologies, spectrum resources are becoming increasingly scarce, and spectrum monitoring technologies targeting control and interference suppression impose higher requirements on the real-time performance, reliability, and intelligence of signal detection, recognition, and key parameter extraction. Traditional signal processing methods heavily rely on operators’ prior knowledge, making it difficult to achieve intelligent spectrum monitoring, and often exhibit poor performance in complex electromagnetic environments with unknown signals or strong interference. Existing deep learning-based automatic modulation recognition techniques are more focused on signal recognition, with relatively limited research on detection and key parameter extraction. To address these challenges, this paper proposes a wideband signal processing frame based on semantic segmentation and signal spectrogram. The frame employs RepViT as the backbone network and achieves detection, recognition, and key parameter extraction of wideband signals through precise semantic segmentation of signal spectrogram. Experimental results on a large-scale synthetic dataset demonstrate that the proposed frame achieves a maximum signal recognition rate (mAcc) of 82.43% and an average signal recognition rate (aAcc) of 65.16% in multi-modulation scenarios and under different noise power levels. In terms of parameter extraction, the normalized root mean squared error (NRMSE) for time parameters (e.g., start time, and duration) is controlled within the ranges of 0.3%−2.8% and 0.4%−1.6%, respectively, while the NRMSE for frequency parameters (e.g., center frequency, and bandwidth) reaches 8.7% and 0.6% in multi-classification tasks, providing an effective reference solution for intelligent wireless signal analysis.
Liu et al. (Mon,) studied this question.