Solar Radio Spectrum Classification by Deep Conventional Neural Network Based on Transfer Learning

The damage to communication facilities and navigation equipments may be caused by the violent solar radio explosion. With the emergence of high-performance observation equipments, solar data shows a tremendous growth trend. Deep learning, especially convolutional neural network, is used to classify solar radio spectrum images because it performs well in the processing of massive data, especially image data. In response to the lack of solar image datasets, a transfer learning method based on four pre-trained models with ImageNet image dataset is improved to solve the problem of classifying and recognizing three categories of solar spectral images. The experimental results show that MobileNet based on this model can achieve 98.95% accuracy, while the training speed of the model is improved also.