| Abstract : | Sunspots, the dark features on the solar surface are actually bipolar magnetic regions tilted by an angle with the equator. The earlier method for the determination of the tilt angle of a sunspot group from a white light image includes a geometric approach. In this method, the centroid of the sunspot group is determined, the spots which are left are taken to be leading poles and assigned positive signs and those right are taken to be trailing poles and assigned negative signs. The positive and negative signs here have nothing to do with their magnetic polarity. The tilt angle obtained by this process is called 'pseudo tilt angle' and can have errors. Another way can be to take the corresponding magnetogram image, look for the particular sunspot group, assign magnetic polarity signs to the regions in that sunspot group and calculate the tilt angle. But good quality magnetograms are only available from 1996 onwards. Thereby comes this supervised machine learning technique, where the machine will generate the 'model', the mapping function, such that when a white light image is given as an input, the model will be able to predict the correct magnetic polarity assigned white light image as output. But the system needs to be trained for the model generation and thus we need training data sets. My work includes the generation of the training data sets only which are white light images, magnetogram images as input, and correctly magnetic polarity assigned white light images as output. |
