Ionogram Echo Extraction Based on the Convolutional Neural Networks

Abstract

AbstractThe extraction of ionospheric echo plays a key role in the research of radio wave propagation theory and is the basis of the design, application, and test of the radio engineering system. However, it is too complicated to achieve because of several negative factors, such as environmental noise, thermal noise, ionospheric time‐varying dispersion characteristics, and so on. In this paper, a novel ionospheric echoes extraction model, referred to as IECAENet, is proposed based on the convolutional neural networks. First, VGGNet is introduced as the classification module to extract echoes from different kinds of ionograms, including vertical ionograms, oblique ionograms, and backscatter ionograms. Then, the echoes extraction module is designed. It considers not only the ionogram features but also an explicit hint, named echo mask, that expresses the location of the echoes. Finally, the residual learning technology and skip connection structure are introduced to improve the performance of the module. Experimental results on a real data set indicate that IECAENet outperforms the baselines for the ionospheric echoes extraction. Using the false alarm probability that IECAENet has an optimal performance as the benchmark, the detection probabilities of the vertical ionogram, oblique ionogram, and backscatter ionogram are improved by 22.18%, 22.56%, and 6.67%, respectively, compared with the traditional methods.