Cloud detection method based on clear sky background under multiple weather conditions

Abstract

This paper deals with an image processing methodology based on a clear sky library for cloud detection. It is a part of a project which aims at using the detection results of ground-based cloud images in combination with solar irradiation data to complete power forecasts for solar power plants at ultra-short-term horizons (15 min). This paper focuses on the implementation of the cloud detection process. To overcome the problem of detection failure caused by uneven background brightness distribution in the sky, detection work is achieved by building a library of clear sky background images for various weather conditions at different solar zenith angles and combining background elimination methods through image rotation and image matching algorithms. A hybrid algorithm is proposed for detection failure when the sun is blocked by the cloud to achieve secondary processing. The test results show that this method has good detection accuracy at the visual level compared to traditional algorithms, especially for situations with low cloud and sky discrepancies such as thin clouds and hazy weather.