A Framework for Multitype Solar Radio Burst Detection and Automated Parameter Extraction

Accurate real-time detection of solar radio bursts (SRBs) is essential for advancing solar research and space-weather forecasting. However, existing methods face several challenges, including limited coverage of SRB types, significant class imbalance, particularly for Types IV and V, insufficient modelling of multiscale and time–frequency features, and the lack of automated parameter extraction. To address these issues, we propose a framework for detecting and analyzing multiple types of SRBs. This framework integrates a denoising diffusion probabilistic model to mitigate sample scarcity and employs the improved You Only Look Once v9, which is capable of capturing time–frequency and multiscale spectral characteristics. Additionally, it includes a parameter extraction method designed to automatically determine start time, end time, and frequency ranges. Experiments conducted on the self-constructed SRB dataset demonstrate that the proposed framework achieves the mean average precision (mAP@0.5) of 0.860, with notable improvements for Types IV (from 0.693 to 0.853) and V (from 0.792 to 0.865), while maintaining a processing speed of 45.4 frames per second. This framework provides an effective solution for SRB detection and parameter extraction, serving as a valuable reference for astronomical data analysis.