Deep Generative Model Fusion Based Industrial Visual Defect Detection System

With the complexity and high precision of the development of manufacturing industry,the defect detection based on traditional digital image processing has been unable to meet the actual needs.The rapid development of Artificial Intelligence has made deep learning-based defect detection methods become one of the mainstream research directions in the field of industrial vision.Deep Learning technology is data-driven and performs defect detection through feature learning of positive and negative samples.The workload of sample labeling is huge,and at the same time,real defect samples cannot be collected enough on the industrial production line,which results in the detection network based on the deep discriminant model cannot fully learn the negative sample features and environmental noise,and the generalization ability is poor.Given these problems,this thesis builds a surface defect detection system for industrial electronic screens based on the deep generative model.The system can alleviate the problem of insufficient training data of the deep model and improve the accuracy of defect detection.The main research contents are as follows:(1)Aiming at the problems of environmental noise and insufficient detection capacity caused by a lack of defective samples in industrial scenes,this thesis proposes a background generative model for industrial vision based on a Deep Convolutional Generative Adversarial Network.Learning the ability to generate texture feature information from positive samples,in the case of using only good industrial pictures.The entire network has a good ability to simulate the diversity of the sample set,and realizes the generation of uneven input pictures caused by complex industrial environments,avoiding the dependence on negative samples and data labeling.(2)Aiming at the problem of insufficient detection accuracy caused by large differences in feature information between different industrial defects,this thesis proposes a background reconstruction algorithm based on the fusion deep generative model,which combines the Auto Encoder with the generative adversarial model in(1).The combination realizes the background reconstruction of the defective picture and transforms the detection problem of Deep Learning into a background reconstruction problem.At the same time,the digital image algorithm is used to complete the reconstruction post-processing and defect positioning tasks to achieve a complete detection system.The research results show that the detection system in this thesis can achieve 89.07%defect detection accuracy under the disadvantages of insufficient negative samples and large environmental noise.