Abstract: When tracking and identifying the bucket target of excavator robot by using visual information, there exist rotation, translation and zoom situations for the bucket images collected in real time. To improve the identifying ability of the bucket target, a recognition method of bucket target was proposed based on invariant moments and BP neural network. The method extacted seven moment characteristic quantities of bucket image with invariant performance against translation, rotation and zoom. They can act as the training and testing samples for improved BP neural network after being normalized. Using the trained neural network to identify bucket target, the simulation result shows that this method has high recognition ability.