Data Cleaning for Photospheric Bright Points Based on Clustering Analysis

Photospheric Bright Points (PBPs) are usually confused with the bright granules near the inter-granular dark lanes, because of their small-scale and fuzzy boundary. This paper uses the K-means and DBSCAN algorithm to differentiate the non-PBPs from PBPs candidates. First, Laplacian and morphological dilatation algorithm is employed to extract PBPs candidates from images, and a three-dimensional algorithm is used for tracking the evolutions of PBPs candidates. Second, seven properties of each candidate are calculated. They are diameter, intensity, eccentricity, the proportion of their boundary in the dark lanes, horizontal velocity, motion type and diffusion index, respectively. After standardizing data, principal component analysis is used for reducing the seven-dimensional data to three-dimensional. At last, non-PBPs are cleaned by K-means algorithm and DBSACN algorithm, respectively. The result shows that both K-means and DBSCAN algorithm can be used to clean the non-PBPs from PBPs candidates. The processing accuracy of K-means algorithm is around 80%, and that of the DBSCAN algorithm is 53%. The result indicates that the K-means algorithm is more suitable for cleaning the non-PBPs than DBSCAN algorithm.