Classification of Whole Sky Infrared Cloud Images Via Sparse Representation

To classify sky conditions automatically basing on cloud images which are obtained from the whole sky infrared cloud measuring system (WSIRCMS), we cast the recognition problem as one classification among sparse representation models and propose a new method using the representation of cloud images from compressed sensing (CS) to address this problem. Firstly, we should construct redundant dictionary with typical cloud samples. Secondly, we should solve the optimal solution of paradigm using Iterative Shrinkage-Thresholding (IST) algorithm. Finally, we should use the residual method to discriminate cloud classification. Using sparse representation theory in cloud classification avoided the feature extraction process, and provided a new way for automatic identification of infrared cloud images. Recognition rates of undulates, stratus, cumulus, cirrus clouds and clear sky were 72%, 92%, 68%, 62% and 98% respectively