Automatic detection and classification of Spread-F in ionograms using convolutional neural network

Abstract

Equatorial spread-F (ESF) is an irregularity caused by plasma instabilities on the night side that causes signal degradation and disruptions to the GNSS signals. Ionosondes could detect ESF as it appears as a diffused echo in the ionogram images. This study proposes a Convolutional Neural Network (CNN) model that can automatically detect ESF within the ionogram images and classify its type. The model has been trained using 2646 manually labeled ionograms from the Low Latitude Ionospheric Sensor Network (LISN) VIPIR Ionosondes in South America. The data used to train the model was measured from 2019 to 2024. The model was able to classify the testing images into six categories: Clear class, frequency spread-F (FSF), range spread-F (RSF), mixed spread-F (MSF), strong spread-F (SSF), and Unidentified class. It demonstrated high classification accuracy within the extracted test subset and a further random test, showcasing robustness and consistency in detection accuracy across all classes. Furthermore, the model performance has been evaluated and compared with other baseline models: VGG16, VGG19, ResNet18, and Inception-V3 in the same environment. Additionally, a comparison with published models is provided. Our model showed a higher consistency in classification accuracy across all classes compared to the mentioned models.