Low Complexity Deep Learning Models for Ionospheric Layer Detection

The ionosphere is a crucial region of Earth's atmosphere for radio communication and satellite navigation. Instruments like ionosondes monitor this region and produce ionograms depicting the ionospheric layers. Although automated ionogram scaling tools exist, they often require manual intervention. Recent advancements in deep learning have shown promise in automating ionogram analysis with improved accuracy. However, challenges persist in detecting several ionospheric layers simultaneously and across diverse datasets. This paper compares a traditional U-net with a less complex Convolutional Neural Network architecture and evaluates the performance on a diverse ionogram database covering various sources, locations and periods within the solar cycle. The proposed method is compared to existing deep learning approaches and shows competitive performance in intersection over union (IoU) and recall metrics.