空间科学学报2025，Vol.45Issue(4)：943-949,7.DOI:10.11728/cjss2025.04.2024-0066
基于声谱图和卷积神经网络的磁暴图像识别
Geomagnetic Storm Image Recognition Based on Spectrogram and Convolutional Neural Network
摘要
Abstract
Geomagnetic storms represent an important type of geomagnetic field disturbance that can cause interference and damage to fields such as communication,power supply,and aerospace technology.Therefore,the advancement and innovation of geomagnetic storm recognition technology have good de-velopment prospects for strengthening the application of geomagnetic storm data in related fields.In this study,we leveraged an extensive dataset comprising minute value recordings of horizontal components sourced from 12 permanent geomagnetic observation stations from 2010 to 2023.Employing spectral imaging technology,we conducted a comprehensive artificial intelligence-based image classification analy-sis to differentiate between geomagnetic storm days and geomagnetic quiet days,utilizing the VGG19 convolutional neural network model.We have obtained good experimental results.This experiment uses accuracy,precision,and recall as evaluation metrics.The experimental model demonstrated the accura-cy rate of 97.41%,with a precision value of 98.00%and a recall rate standing at 96.80%.These indica-tors collectively emphasize the reliable predictive ability of our model.Furthermore,the application of spectrograms within the context of image recognition and classification has demonstrated significant fea-sibility.Notably,the VGG19 convolutional neural network model exhibited remarkable feasibility when tasked with categorizing geomagnetic storm days and geomagnetic quiet days.The recognition accuracy of this model for geomagnetic storm days is relatively high and the model itself is relatively stable.How-ever,there is some fluctuation in the recognition of geomagnetic quiet days,which also means that the model still has room for further improvement,especially by increasing the number of training sets and improving the learning accuracy of the model for map information.In summary,our research findings contribute to the improvement of geomagnetic storm identification methods,providing a promising ap-proach to enhance geomagnetic storm prediction and monitoring capabilities,and ultimately promoting the wider application of geomagnetic storm information in related fields.关键词
地磁/磁暴/声谱图/卷积神经网络/图像分类Key words
Geomagnetism/Geomagnetic storm/Spectrograms/Convolutional neural networks/Image classification分类
天文与地球科学引用本文复制引用
李鸿宇,孙君嵩,王丽,杨杰,赵雨馨..基于声谱图和卷积神经网络的磁暴图像识别[J].空间科学学报,2025,45(4):943-949,7.基金项目
中国地震局地震科技星火计划项目(XH23016YB),江苏省地震局青年科学基金项目(202408)和江苏省地震局局长科研基金项目(202201)共同资助 （XH23016YB）
