Model learning method, training method and system for industrial visual inspection

The invention discloses a model learning method, training method and system for industrial visual inspection. The method is applied to a convolutional neural algorithm, and comprises the steps of collecting a training set and a test set of a detection target, obtaining an open source data set, and obtaining a trained base class target detection network model; defining a base class and a new class according to the task description, and constructing a first full-class training set containing all classes and a second full-class test set containing each class according to the sample quantity of the new class, wherein the sample quantity of each class is K; modifying an RPN network of the base class target detection algorithm model, and obtaining full class features through the modified base class target detection algorithm model; and decoupling the modified base class target detection algorithm model. According to the process, the requirement for the number of training samples can be remarkably reduced, and the r