Image Classification of Solar Radio Spectrum Based on Deep Learning

Aiming at the problems that traditional image denoising methods cannot well filter the background noise of solar radio spectrum images, and the training sample data is small and unbalanced, it is proposed to segment the features and background of the spectrum image through Gaussian filtering and image binarization. Then use the morphological closed operation to enhance the feature; through the combination of image transformation and random indexing, the problem of uneven distribution of various samples in the solar radio spectrum database is solved. By designing the structure of the convolutional neural network, the classification model can better extract image features and improve the classification accuracy. Experimental results show that the convolutional neural network combined with image preprocessing has achieved an average TPR value of 97.97% on the solar radio spectrum data set, which is better than the existing research results and has application value for the automatic classification of solar radio spectrum images.