According to the imaging features of dim targets, we put forward a target detection algorithm based on the local maximum mean of neighborhood and multi-scale morphological filtering.It is determined whether the image center is the maximum or not by the sliding window.If it is the maximum, the center pixel will be replaced by the maximum mean of the two directions of its four neighborhoods.Otherwise, the center will be assigned according to the calculated weighting coefficient by figuring out the two maximum gradients of four neighborhoods.It is implemented through the whole image for eliminating the noise and improving the signal-to-noise ratio of the original image.Then, multi-scale morphological filtering of the image is carried out and the background can be estimated effectively and be removed from the original image.After the threshold is calculated by using the improved adaptive segmentation method, the targets are extracted from candidate points.Multi-frame association is applied to sequential images for further reducing the false alarm rate.Experiments show that, the method is easy to implement, and can detect the whole target with high detection probability and low false alarm rate.