Solar Radio Spectrum Image Classification Based On Deep Learning

Solar eruptions have a serious impact on the Earth and even disrupt communication equipment and navigation systems.At the same time,there are many radio phenomena,which can feed back a large number of parameters closely related to solar activity,so solar radio research has a broad prospect of scientific research and application.Due to the investment of high-performance observation equipment,the huge growth of the image data of solar radio spectrum has put forward higher requirements for automatic data analysis.With the development of in-depth learning,the feature extraction and classification of solar radio spectrum image using in-depth learning technology has been widely studied.In this paper,the existing classification methods of solar radio are introduced.A variety of image processing techniques are used to pre-process the solar radio spectrum data.In view of the characteristics of the solar radio spectrum image data,depth learning is applied to the classification task.Firstly,for the data in the solar radio spectrum image database,we preprocess them by means of scale transformation and normalization,Using sample amplification technology to solve the problem of sample imbalance.On this basis,the convolution neural network is used to effectively improve the classification effect of solar radio spectrum images by optimizing the network structure.Secondly,we use image segmentation,image denoising and image enhancement to process the spectrum images of solar radio bursts for type III and type IV bursts,it makes the features of burst spectrum image more obvious and facilitates the training of subsequent models.Then we use the convolution neural network model to achieve 95.35% classification accuracy;And it achieves better image recognition effect than the ordinary convolution network under the condition of little time difference,which makes the recognition rate reach 96.91%.The experimental results show that this method can effectively improve the classification effect of solar eruption types.