A Deep Learning-Enabled Ionogram Dataset for Detection and Classification of Low-latitude Spread-F Phenomena

Research on the ionospheric Spread-F (SF) phenomenon holds significant value in both fields as ionospheric electrodynamic research and enhanced operational applications in radio-based technologies (e.g., communication and navigation). To date, the classification of Spread-F remains largely reliant on the manual interpretation of ionograms by experts, suffering from inefficiency (~10 seconds per figure) and subjectivity. There has been no publicly available ionogram dataset classifying Frequency/Range/Mix/Strong Range SF (FSF/RSF/MSF/SSF) by either human labor or machine processing. To address this problem, we introduce the first open, expert-guided ionogram dataset that is simultaneously the most comprehensive in terms of class coverage, the largest in volume, and the most extensive in temporal span. This collection encompasses 150,000 ionograms (30,000 per class, including a "non-SF" group) spanning 14 years from 2002 to 2016, thereby capturing a diverse range of solar and geomagnetic conditions. The attached classification SA-ResNet50 model based on this dataset could be applied to further data.