Deep learning-based sample generation method in industrial vision detection

Abstract

The invention relates to a deep learning-based sample generation method in industrial vision detection, which aims to solve the problems of few existing vision detection samples, long sample trainingtime and high similarity among the samples, and comprises the following steps of: a) extracting defects in a real defect picture; b) fusing the extracted defect with a defect-free picture so as to generate a pseudo defect picture; c) processing the pseudo defect picture and sending the processed pseudo defect picture to a generation network in the adversarial neural network to generate a defect picture; d) continuing to generate a pseudo defect picture according to the step b), and then iteratively training the network repeatedly according to the step c); and e) when the training in the step d) meets the requirements, adding the generated defect pictures into the sequence of the real defect pictures, and continuing iterative training. Finite samples can be utilized to generate a large number of samples meeting the requirements, the training time is shortened, and meanwhile, the similarity among the samples is low.

Description

technical field [0001] The invention relates to a sample generation method, in particular to a sample generation method based on deep learning in industrial visual inspection. Background technique [0002] With the opening of the era of artificial intelligence, deep learning, as an advanced technology, has been widely used in various fields. In the field of machine vision, the detection effect and ease of operation of algorithm software based on deep learning are far better than traditional methods, but there are still the following problems: [0003] 1. Deep learning requires a large number of sample images to train the neural network, but considering the actual situation, there may be a lack of defect samples due to low product defect rates and uneven distribution of defect types. [0004] 2. At present, there is an attempt to use an adversarial neural network to generate sample images to solve the above problems, but the existing adversarial neural network generation def...

AI Technical Summary

Problems solved by technology

[0006] The technical problem to be solved by the present invention is: in order to solve the problems of few samples in the existing industrial visual inspection, long training time for the samples generated by conventional adversarial neural networks, and high similarity between the generated samples: to provide an industrial visual inspection Sample Generation Method Based on Deep Learning

Images