In order to realize the accurate detection of small dim target in the sea through infrared image,a detection algorithm based on local peak detection and pipeline-filtering are pu t forward by my team.The algorithm first extracts some suspected targets by local peak detection,and then removes most of the non-target peaks according to the self-adaptive threshold processing.Finally,the residual inte rference is eliminated by the pipeline filtering method to identify the target accurately.In view of the characteristi cs of the algorithm including a large number of conditions and parallel computing,this paper compares the work charac teristics of CPU and GPU,and finally speeds up the algorithm by using the heterogeneous computing model of CP U-GPU collaboration.The experimental results show that the leakage alarm rate of the target detection is not higher than 3.5%,the false alarm rate is lower than 5%,the acceleration ratio is 26,the rate of resolut ion processing 640×512images is not less than 32frames per second,and it has high engineering application value.