Abstract: For the purpose of discrimination between cloud area and earth object in satellite remote sensing image, a pattern recognition approach comprising multi-feature extraction and minimum distance classification is proposed. For fundamental image block, its distinguishing rate is more than 90% with low computational complexity. The deletion rule in frame-partition is determined to remove large volumes of cloud area from the image data. Combining parallel DSP architecture, the proposed algorithm can fulfill the tasks of cloud detection and removal in real-time. The equipment utilizing above approach has been installed in a satellite ground station. Its performance is satisfied with the technical requirements. The information processing capacity and the level of automation of the satellite ground station are improved dramatically.