Cloud Types Identification for Meteorological Satellite Image Using Multiple Sparse Representation Classifiers via Decision Fusion

Meteorological satellite can monitor the weather conditions in large scales effectively; some solutions and researches have been raised for cloud types identification in satellite cloud image analysis. Extracting the features of the satellite image and designing effective classifier play important roles in implementing cloud types identification system. Since different features describe the characteristics of the cloud image in different perspectives, the collaborative utilization of the different features help to improve the accuracy of cloud classification. This paper proposed a new method to identify cloud types from meteorological satellite image using multiple sparse representation classifiers via decision fusion. First, followed by different types of features extracting, multiple sparse representation-based classifiers were trained respectively. Then, the strategy of decision fusion was introduced to fuse the outputs of multiple classifiers. In order to bring about a reasonable fusion rule, the fusion weights were determined by an adaptive iterative procedure, and the iterative procedure was constructed according to the performance of each sub-classifier. Finally, an adaptive weighted fusion was implemented to determine the cloud type according to the outputs of sub-classifiers and their corresponding weights. The experimental results on FY-2G satellite data demonstrate that the proposed method gains higher recognition accuracy than each separated sub-classifier, which suggests that the strategy of decision fusion can take advantage of each sub-classifier. Moreover, the proposed method achieves competitive results when compared with the other state-of-the-art methods. The computation efficiency of the proposed method is also analyzed briefly.