A method for cloud detection and opacity classification based on ground based sky imagery

摘要: Abstract. Digital images of the sky obtained using a total imager (TSI) are classified pixel by into clear sky, optically thin and thick clouds. A new classification algorithm was developed that compares red-blue ratio (RBR) to RBR library (CSL) generated from captured on days. The difference, rather than ratio, between CSL resulted in more accurate cloud classification. High correlation TSI image aerosol optical depth (AOD) measured an AERONET photometer observed motivated addition haze correction factor (HCF) model account for variations AOD. Thresholds clouds were chosen based training set validated with manually annotated images. Misclassifications opposite category less 1%. Thin accuracy 60%. Accurate detection opacity techniques will improve short-term solar power forecasting.