An improved dynamic programming method based on position predication was proposed in the paper to quickly detect dim and small debris from the space-based observed images on GEO band. Firstly, after analyzing the observed data, we found that different GEO debris have same movement velocity in images, after further data fitting, we obtained a mapping model between the velocity of image debrits and sub-star latitude of observation satellite and proposed a method to estimate the debris velocity in space-based observation images on the GEO band, at last, debris position was predicted by estimating velocity,then position information weight was applied in traditional dynamic programming recursive equation, and the object searching range was obtained based on position prediction,resulting in the number of object status in the recursive equation were reduced. Measured data verified that fitting deviation of velocity mapping model was within 1pixel. We selected the measured data within typical observation period for test and the results show that the detection time taken by proposed method has been reduced by more than 90% as compared with traditional dynamic programming methods, and the virtual scenery rate has also been reduced by more than 5.9%.It can be conclusion that the method is suitable for detection of dim and small debris from the space-based observation images on GEO band.