Automated classification of sunspot groups with support vector machines

Abstract

A new effective technique is presented for automatic classification of sunspot groups on full disk white light (WL) solar images. This technique is implemented on images taken from the Solar Oscillations Investigation Michelson Doppler image (SOI/MDI) aboard the Solar Heliospheric observatory (SOHO). The technique focuses on employing Support Vector Machines (SVMs) as effective classification tool. In addition to applying SVMs the problem of extracting sunspots and sunspot groups from solar image is solved in an efficient and different way from the ones previously employed. This technique proceeds in several consequence phases. The first phase involves solar disk image extracting. The second phase involves Binarization and smoothing of an extracted solar image disk. The third phase involves unsupervised segmentation of sunspots groups. This phase consists of two subphases: a) extracting the spots and, b) combining together the sunspots that belong to the same group. The fourth phase involves attributes extraction of each sunspots group. The final phase is the classification phase using SVMs, by applying a one-against-all technique to classify the sunspots groups. The proposed technique has been tested using different sets of sunspots achieving approximately 88.56% recognition rate for Zurich classification of sunspot groups. © Springer Science+Business Media B.V. 2010.