The presence of clouds affects a wide application of remote-sensing images. Based on the high spatial resolution and wide band range of the visual and infrared multispectral imager (VIMI) of the hyper-spectral observation satellite, an improved multichannel-threshold cloud-detection algorithm applicable to the VIMI data is proposed. First, potential cloud pixels and clear pixels are separated according to various characteristics of clouds in the visible-thermal infrared channels. Then, the probabilities of temperature, variability, and brightness are combined to generate a cloud mask over land and water. Finally, clear-sky restoration tests are applied to the potential cloud layers to reduce the misclassification of clouds over land, water, and snow/ice scenes. The results of the improved multichannel-threshold cloud-detection algorithm are quantitatively compared with those of the conventional cloud-detection algorithm. The results show that the improved algorithm can be applied to different surface scenes to obtain better detection results with an average overall accuracy of 92.0%, and the overall difference is reduced by 3%. Furthermore, the average cloud-pixel accuracy and clear-sky pixel accuracy are obtained as 92.4% and 91.8%, respectively. The results show substantial reduction in misclassification and omission errors; especially, on the bright surface, the average cloud-pixel accuracy over a city and snow surfaces improves by 4% and 5% and the difference decreases by 4% and 2%, respectively. The improved cloud-detection algorithm outperforms the conventional algorithm in terms of high efficiency operation.