Provided in the present invention is a method for automated detection of defects in cast wheel products, comprising three stages of sample preprocessing, offline training, and online inspection. Specific steps comprise collection of training samples; sample preprocessing; categorization of samples in three major types as spokes, rims and hubs, then processing; by means of the three sample types, spokes, rims and hubs, training a network offline, then generating online detectors for said three sample types respectively; importing the trained CNN spoke defect detector, CNN rim defect detector and CNN hub defect detector to a host computer, and locating same on an automated production inspection line; wholly automated online defect detection; output of identified detected defects by an output mode according to user requirements. The automated inspection method of the present invention is particularly robust in respect of the illumination, angle of photography and positioning that prevail during the inspection pr
