Cloud Detection on Ground-Based Sky Images with Brightness Reduction of Circumsolar Region

Cloud detection is an essential process to extract cloud information for solar irradiance forecast. This process is generally applied to sky images to identify cloud and sky pixels. The glare in circumsolar region is a main obstacle for cloud detection. Pixels are incorrectly identified when using basic thresholding techniques. Thus, we propose an algorithm with brightness reduction of circumsolar region to deal with this issue. Non-uniform illumination of sky images is corrected by using homomorphic filtering, then sky conditions of images are categorized. Cloudy and overcast images use normalized blue/red ratio (NBRR) thresholding for identifying cloud pixels, whereas clear sky and partly cloudy images use thresholding technique. We take 100 images to evaluate the performance of the proposed algorithm. The results show that our algorithm outperforms basic NBRR method with accuracy improvement of 6% and 5.3% for clear sky and partly cloudy images, respectively.