Classification of clouds in satellite imagery using over-complete dictionary via sparse representation

Abstract

Satellite imagery provides distribution information of cloud in a wide range of spatial and temporal scales. To improve the accuracy of weather forecasting and enhancing the effectiveness of climate monitoring, it is important to study the classification of clouds in satellite imagery. Motivated by the concept addressed in sparse representation using over-complete dictionary, a novel method for cloud data classification in satellite imagery was proposed. The intensity and spectral features attributed to different cloud types of samples were used to construct an adaptive over-complete dictionary in order to represent the cloud samples sparsely, followed by extracting the dictionary feature via sparse coding of samples. Then, the sparse representation coefficients matrix of training sample sets for specific cloud type was used to form a projection subspace. After orthonormal processing for the projection axis of the projection subspace, an effective sparse classifier was designed. Finally, the problem of cloud type identification for the test sample was solved according to the similarity between the test sample and specific cloud type subspace. Experiments were conducted on meteorological satellite data, the promising results indicated that the proposed approach can identify clear water (CW), clear land (CL), cumulonimbus (CB), altostratus & altocumulus (AS&AC), cirrostratus & cirrus-densus (CS&CD) and nimbostratus & cumulus (NS&CU) in the satellite images effectively, the classification accuracy is higher than that of the support vector machine classifier and traditional sparse representation classification classifier.