Defect Detection Of Industrial Parts Based On Deep Learning

In the production process of industrial products,due to aging production equipment,unqualified raw materials and backward production technology,some products appear defects,which may not meet the functional characteristics required by the product.In order to ensure the qualified rate and quality of products,defect detection is a crucial step.Manual detection and traditional machine learning methods have many drawbacks,such as high cost,low efficiency and poor stability.With the continuous development of computer technology,more and more technologies such as deep learning and computer vision are applied in the industrial field,but at present,most methods only use the Convolutional Neural Network(CNN)structure,which cannot make good use of the global information.In addition,most of the current methods are based on supervised learning,and the datasets of industrial product defects are often seriously unbalanced,and in extreme cases,there are only defect-free data,in which it is difficult to detect defects based on supervised learning.Based on this situation,two defect detection algorithms based on deep learning are proposed in this paper for defect detection in industrial scenarios.The research contents mainly include the following two points:(1)This paper proposes a supervised defect detection algorithm based on deep learning.The algorithm fuses the features extracted by CNN and Transformer through the fusion module,which combines the advantages of both,so that the model can not only focus on local information but also capture remote dependencies well,which solves the problem that most of the current algorithms cannot make good use of global information,and shows good performance in the experiment.(2)This paper proposes an unsupervised defect detection algorithm based on deep learning.Based on the idea of image reconstruction,the algorithm uses MAE to reconstruct image blocks at defects,repair image blocks into defect-free forms,and locates defects by comparing the structure similarity of images before and after reconstruction.The algorithm is implemented based on unsupervised learning,without the use of labeled defect data,and achieves defect detection in the scenario of severely unbalanced datasets,and shows good performance in the experiment.