改进的主成分分析网络极光图像分类方法

Improved PCANet for aurora images classification

极光的不同形态蕴含了不同的物理意义,进行极光图像的分类研究对人类生活具有极其重要的科学价值.笔者在简单的深度学习模型主成分分析网络的基础上提出了一种改进的主成分分析网络极光图像分类方法.首先利用改进的主成分分析网络提取极光图像的特征,然后将所得特征输入支持向量机对极光图像进行分类.在中国北极黄河站的全天空图像数据库的分类实验结果表明,所提方法取得了较高分类准确率.

The mysterious aurora is changeable, and the different forms of the aurora represent various physical processes which often affect our lives. So, it is of significant scientific value to classify the aurora images for the study of space physics. Based on the PCANet, a simple deep learning model, we develop an improved PCANet algorithm for aurora images classification. Firstly, the map of aurora images are extracted by the improved PCANet. Then the support vector machine is used to classify the feature of aurora images. Experimental results with the dataset obtained from the All-sky Imager at the Chinese Arctic Yellow River Station demonstrate that the scheme can obtain higher accuracy in aurora image classification than the PCANet.