H-Alpha anomalyzer: An anomaly detector for H-Alpha solar observations using a grid-based approach

Abstract

This article presents a Python package named H-Alpha Anomalyzer for detecting anomalous H-Alpha observations of the Sun. Using this open-source package, users can transform the labor-intensive task of filtering anomalous observations from millions of instances, thereby enhancing the quality of data used for data-hungry algorithms, particularly Deep Neural Networks (DNNs). Our region-based probabilistic method offers explainability by assigning anomaly likelihoods to each cell of a given observation. Additionally, users can set a probability threshold to customize the degree of anomaly required for an entire image to be classified as anomalous. This paper also reports the quantitative validation of the method. On a modest laptop computer, this lightweight package processes ten 2k-by-2k-pixel images per second, which is significantly faster than its DNN-based counterparts.