Fixed pattern noise removal for solar images using a Self-Supervised Destriping Network

Le résumé fourni par la source

With the development of Complementary metal oxide semi-conductor (CMOS) fabrication process, CMOS Image Sensor (CIS) is widely used in solar high-resolution observations. However, the images often suffer from obvious fixed pattern noise (FPN). In this paper, we present a innovative technique framework for the FPN correction. Firstly, the noise-free image is roughly estimated by the 1D Guided filtering method and used to make up the training pairs. Secondly, a Multi-scale Attention Convolution Neural Network (MA-CNN) is proposed to achieve single image correction. In the neural network, a Multi-scale Convolution (MS-Conv) unit is designed to extract complementary FPN features at different scales. Thirdly, the channel attention mechanism is added to the network, which can more accurately separate the FPN from the image. Finally, the visual effect and quantitative assessment demonstrate that the proposed method has a better performance compared with the traditional methods.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.