[PEM12-P08] Automatic detection of Spread-F on ionogram using Machine Learning
Keywords:Ionosphere, spread-F, plasma bubble
Ionospheric irregularities referred to as equatorial spread-F are very important phenomena in terms of radio wave propagation because their spatial scales are from centimeters to tens of kilometer and they affect wide-band radio waves. Therefore, they interfere the reliability of satellite-ground communications, navigation systems and so on. The ionozonde is one of the useful observation systems of spread-F. If spread-F in the ionogram, which shows the reflection height of HF radio waves versus carrier frequency, is detected automatically by a machine learning system, real time detection of spread-F could be replaced by the automatic detection system using the machine learning system. In this study, therefore, we have developed an automatic detection of spread-F in the ionogram using machine learning.
In order to construct the classifer, the ionograms observed at 4 observatories (Wakkanai, Kokubunji, Yamagawa, Okinawa) utilized by National Institute of Information and Communications Technology were prepared. As for learning data, we used 400 ionograms with spread-F observed at each stations in 2018. HOG (Histogram of Oriented Gradients) feature extraction is used in obtaining the feature quantity.
Using the classifier to detect spread-F in the ionograms observed at Okinawa in May 2019, a detection rate reached about 90%. This classifier is also enable to detect the spread-F in the ionograms obtained at Kokubunji and Yamagawa with the detection rate of 90%. However, the detection rate of the spread-F in the ionograms observed at Wakkannai is very low (about 60%). In the presentation, we will present the comparison with another observation point at the same period.
In order to construct the classifer, the ionograms observed at 4 observatories (Wakkanai, Kokubunji, Yamagawa, Okinawa) utilized by National Institute of Information and Communications Technology were prepared. As for learning data, we used 400 ionograms with spread-F observed at each stations in 2018. HOG (Histogram of Oriented Gradients) feature extraction is used in obtaining the feature quantity.
Using the classifier to detect spread-F in the ionograms observed at Okinawa in May 2019, a detection rate reached about 90%. This classifier is also enable to detect the spread-F in the ionograms obtained at Kokubunji and Yamagawa with the detection rate of 90%. However, the detection rate of the spread-F in the ionograms observed at Wakkannai is very low (about 60%). In the presentation, we will present the comparison with another observation point at the same period.