Defect Detection Method Based On Generative Adversarial Network

In recent years,machine learning and computer vision inspection methods have been widely used in the field of defect detection.Since the surface defects of industrial products will adversely affect the aesthetics,comfort,and performance of the products,the detection of product surface defects is particularly important for manufacturing enterprises.The use of machine learning detection methods can largely overcome the shortcomings of manual detection methods,such as the low accuracy and poor real-time problems of manual sampling.Due to the scarce number of defective samples,sample collection often faces the problem of small sample size and sample category imbalance,resulting in most existing machine learning detection methods cannot obtain good accuracy and speed in industrial product testing.For industrial product defect detection under actual conditions,it is necessary to automatically and accurately detect defects in a small number of samples.Deep learning has a strong ability to learn the basic characteristics of the data set,which can meet the requirements of high precision and high speed at the same time.But training small-scale datasets with high-precision deep learning models is still a challenge,and generative adversarial networks can solve small samples and sample imbalances well.As a variant of the generative adversarial network,the deep convolutional generative adversarial network successfully extended the application data space of the generative adversarial network from the continuous data space to the discrete data space.This research provides a possibility for the application research of generative adversarial networks in small samples and imbalance samples in industrial product defect detection,and also provides new ideas for neural network optimization and other issues.A common problem in defect detection is the inconsistent shape and size of sample defects,and the uneven distribution of categories.To solve this problem,we propose a method that combines transfer learning and generative adversarial networks.The specific work content is as follows:(1)In order to solve the problem of sample imbalance,the original loss function of the VGG16 deep learning network is replaced by the Focal loss function,so that the improved VGG16 network is more suitable for small,imbalanced sample data sets.(2)In order to further improve the performance of the algorithm,this paper proposes to use deep convolutional generative adversarial networks to expand rare class samples to balance the class distribution.After expanding the original data set,the situation of uneven sample distribution was alleviated to a certain extent.(3)The improved defect detection algorithm based on the generative adversarial network is applied to the field of defect detection of actual industrial products.In the process of drug production,quality defects in bottle caps often exist,which affects the quality of the drug.In order to solve this problem,the algorithm in this paper is applied to the defect detection of bottle caps in vials to detect defective vials,and has achieved robustness in accuracy and speed.In addition,this paper compares the proposed defect detection algorithm based on generative adversarial networks with existing machine learning detection algorithms through experiments.The experimental results show that the defect detection algorithm based on the generative adversarial network can obtain high precision and high speed in the industrial application field,that is,the defect detection of the vial,and it can also get better results in detecting abnormal abnormal pictures in the public data set.result.