基于改进的卷积神经网络的太阳射电频谱图像分类

Image Classification of Solar Radio Spectrum Based on Improved Convolutional Neural Network

剧烈的太阳射电爆发可能破坏通信设施和导航设备.随着高性能观测设备的投入,太阳数据呈现海量增长趋势,因此如何应用深度学习技术进行太阳射电频谱图像研究日益得到关注.实验在应用图像处理技术对图像进行处理后,通过调整卷积核和激活函数对卷积神经网络进行改进,实现对太阳射电爆发中Ⅲ、Ⅳ和其他类型的分类.实验结果表明在应用多层卷积神经网络进行分类时激活函数采用ELU函数比用ReLU函数更稳定.

Violent solar bursts can damage communications and navigation equipments. With the investment of high-performance observation equipments,solar data presents a massive growth. Therefore,it has become more popular how to apply deep learning on the research of solar radio spectrum images.After the image processing by experiment applied image processing technology,the convolutional neural network is improved by adjusting the convolution kernel and activation function to realize the classification of Ⅲ,Ⅳ and other types of solar radio bursts. Experimental results show that taking ELU as function is more stable than ReLU in the multi-layer convolutional neural network.