Gaussian Mixture Model Based Cloud Detection for Chinese High Resolution Satellite Imagery

Domestic earth observation satellites, such as ZY-3 and GF-1, have few image bands and limited spectral range. Focusing on those features, this paper outlines a cloud detection algorithm that automatically obtains the grey threshold through histogram fitting by the Gaussian Mixture Model. The algorithm obtains initial parameters automatically from the image histogram; and then it adjusts these parameters by iteration based on the Expectation Maximum principle. It automatically obtains the grey threshold between cloud and clear sky in a chosen image, in line with the distribution features of the components in the Gaussian Mixture Model. Experimental results show that this method has strong advantages, as the range of the spectral bands are not a limit and thus is suitable for both cloudy and cloudless images. In addition, the proposed method needs no auxillary information or manual intervention. It has high computational accuracy and efficiency meeting the requirements for automatic engineering production.