Abstract: This paper presents a revised type of Asymmetric AdaBoost algorithm—M-Asy AdaBoost. It can ensure the success of training process by sample weight distribution. The weight of classifier is optimized by adopting error rates of positive sample. The capability of recognizing positive sample is enhanced. The detection probability is improved and increased monotonously by restricting the classifier adding in the ensemble. The proposed algorithm can attain high detection rate with low false alarm ratio. In the end, it is proved by computer simulation that the M-Asy AdaBoost is effective.