In order to remove the influence of thin cloud on satellite image effectively, an algorithm based on Mallat wavelet transform is proposed. It’s possible to decompose the image into high frequency detail and low frequency approximation components. Based on the facts that cloud noise occupies lower frequency part in the distribution characteristics while scenery information make up the relative high part, this kind of algorithm processes cloudy zones by using linear methods according to cloud thickness in the largest scale of low frequency sub-band image. Different scales of high frequency sub-band is enhanced by non-linear enhancing operators in order to improve images’ sharpness and reduce the impact of residual cloud. Later median filtering is added to process the reconstruct image to reduce the influence of high frequency mutation cloud. The algorithm is utilized to process GF-1 images. Experiment shows that this kind of algorithm can remove thin cloud while it can also preserve image details and edges the same time, which indicates it is better than traditional wavelet transform method.