Knowledge discovery based on statistics of MDI Magnetogram

Solar flare is one of the most powerful activities, which plays a very important role in daily life and space weather, so it is meaningful to predict solar flare accurately. Now expert system is also the main prediction method, for the nature mechanism of solar flare is unclearly, so prediction is often inaccurate. As the increasing of human expectation to discover the outer space, more and more satellites were sent, especially around the sun, so the number of solar daily observed images is more than 1 TBs and with a very high resolution, so it is particularly important to search solar activities from image information. 11 years of SOHO/MDI Magnetograms are selected which contain more than 70 thousands samples. 35 image features are extracted, three of them are usually used by physicists and others are information from image. Three evaluation methods are used to measure the single feature performance and their relationships, which imply that: 1.Length of neutral line is the most powerful predictor; 2. Predictors of morphology evolution of active regions are all very useful and better than the maximum horizontal gradient which physicists used, but with the analysis of relationships between each other, they all have very strong correlation; 3. Although magnetic field distorted portrayals are less power with solar flare, it play very good quality information complementary with the other better features. Through the statistics of large-scale samples, the relationship of solar image features is more clearly and a more comprehensive collation result of relationship between active region and solar flare is got.