Automatic Detection and Classification of Spread‐F From Ionosonde at Hainan With Image‐Based Deep Learning Method

Abstract

An intelligent Spread‐F image detection and classification method is presented in this paper based on an ionogram image set using deep learning models. The ionogram images from the Hainan station, spanning from 2002 to 2015, have been manually labeled into five categories, resulting in a unique ionogram image set for supervised learning models. To balance the number of different types, simulated noises were added to these images. Based on 80,000 samples with Spread‐F and 20,000 samples without, numerous experiments have been conducted to train VGG, ResNet, EfficientNet, ViT, MobileNet, and other networks. The results on the test set indicate that these models except VGG have a good ability of exacting features of different types, leading to a high level of accuracy in detecting Spread‐F and a relatively accurate classification of it. The ionogram images in 2016 are then employed as another test set to further examine the performance of the trained models. Both quantitative and qualitative analyses have demonstrated the results obtained by deep learning models are highly consistent with manual identification.

Prediction of Ionograms With/Without Spread‐F at Hainan by a Combined Spatio‐Temporal Neural Network

“…Table 3 presents an example of a confusion matrix for the evaluation of ConvGRU models with six blocks and a kernel size of 7*7 (referred to as 6‐blocks 7*7, as depicted in Tables 4 and 6), based on data solely from the year 2016 (not included in the training data). The classification results of authentic ionograms of 2016 by a deep learning method (Wang et al., 2023) are taken as the Ground Truth (GT) values. The values in each column, except “ALL,” are the classification results of the second predicted ionograms by our method using the same deep learning method.…”