Application of a Neural Network for Identifying Erroneous Solar Images

For over a century, solar images have been captured across different spectral ranges. Initially, these images were taken on photographic plates, and with the development of CCD cameras, the images transitioned from analogue to digital formats. Analyzing digital images enables us to identify and analyze trends and features on the solar disk more efficiently. However, complications due to instrument malfunction or environmental factors can result in suboptimal images. Traditionally, several statistical parameters are used to check image quality, but these measures do not always yield satisfactory results. In this article, we describe a convolutional classification neural network for near-real time image quality assessment of GONG Dopplergrams. We also present a case study where this approach significantly improved the quality of science data products in an automated data reduction pipeline without any human intervention.