Abstract: Learning aims to emphasize effective stimulation of plastic structure and to form steady memory mode.Data-based machine leaning is an important avenue to make information system attract external information.It is a key technique in field of artificial intelligence.Based on a practical neural network model of filtering for opening and closing,an optimal training algorithm for image target detection is proposed.By means of combining guidable method with the training rule,an adaptive BP training algorithm to adjust learning parameter dynamically is adopted.Experimental results show that the algorithm has invariant property with respect to shift,scale and rotation of moving target in continuing detection of moving targets.