To solve the problem of the inapplicability of the existing cloud detec tion algorithm in the cloud detection of panchromatic remote sensing images at sub-meter level ,a large scale adaptive matching threshold (LS-AMTH) algorithm is proposed in this paper.In the algori thm,the feature set composed of spectrum,texture and edge is structured,and the sub-blocks of images in large scale are classified by using the boost algorithm first.We can obtain a series of cloud region images (some images may include less ground regions) and ground region images (some images may include less cloud regions) through this step.In order to detect the ground region pixels in cloud region images and cloud region pixels in ground region images,t he classified sub-blocks are selected by adaptive threshold matching respectively next.The c loud detection in pixel-scale is obtained and the ratio of cloud is computed at last.The resul ts show that the algorithm has better performance than traditional algorithms,the accuracy o f this algorithm is no less than 97.3% with regard to panchromatic remote sensin g images at sub-meter level and the detection results in complicated cloud-gro und mixed regions are fine.