Cloud detection is important for providing necessary information such ascloud cover in many applications. Existing cloud detection methods includered-to-blue ratio thresholding and other classification-based techniques. Inthis paper, we propose to perform cloud detection using supervised learningtechniques with multi-resolution features. One of the major contributions ofthis work is that the features are extracted from local image patches withdifferent sizes to include local structure and multi-resolution information.The cloud models are learned through the training process. We considerclassifiers including random forest, support vector machine, and Bayesianclassifier. To take advantage of the clues provided by multiple classifiersand various levels of patch sizes, we employ a voting scheme to combine theresults to further increase the detection accuracy. In the experiments, wehave shown that the proposed method can distinguish cloud and non-cloudpixels more accurately compared with existing works.
展开▼