Evaluation on Learning Strategies for Multimodal Ground-Based Cloud Recognition

As a sign of atmospheric processes, clouds play a crucial role in regulating the earth energy balance, redistributing surplus heat and hydrologic cycle. Appropriate recognition method is essential for accurate ground-based cloud classification. This paper evaluates three kinds of learning strategies, i.e., end-to-end method, k-nearest neighbor (KNN) classifier, support vector machine (SVM) for multimodal ground-based cloud recognition. The experimental results demonstrates that SVM is superior to the other methods on multimodal ground-based cloud recognition.