In order to avoid the influence of cloud obstruction on the optical satellites to obtain effective ground target information during the remote sensing satellite imaging process, this paper proposes an on-board cloud detection and avoidance algorithm. Firstly, the adaptive threshold is used to segment the target and background, and the segmented binary image is marked. Then the concept of shape complexity is proposed and the shape complexity of each connected component is calculated. After setting the threshold, the cloud region is extracted. Then, the satellite orbital space coordinate system and the model of avoiding clouds are established. The formulas for adjusting the attitude of agile satellites are deduced, and the satellite network is designed to observe the targets through the intersatellite communication in the visible window. Through experiments and simulations, the time for detecting clouds by utilizing the algorithm is in the range of 168-281 ms, and the correct rate is 89%. The satellite data utilization rate is increased by 18.15% and 22.21% respectively under the two sets of orbital parameters. Basically, it can effectively combine the accuracy and realtime indicators, and alleviate the pressure on the transmission channel of the massive data of remote sensing images.