Research On Panel Surface Defect Detection Based On Deep Learning

During the production process of products,due to the influence of manufacturing processes and complex production environments,it is inevitable that some products have defects on their surface.Therefore,in order to ensure the appearance quality of products,it is particularly important to detect defects on their surface.In recent years,deep learning algorithms have achieved excellent performance in the field of vision,gradually replacing traditional algorithms as the mainstream product surface defect detection method.However,the collected defect images often require improvement in the algorithm performance of existing defect detection networks due to problems such as a small number of samples and a small area of defect areas.Therefore,this article focuses on the research of deep learning algorithms for panel surface defect detection,and the specific work includes the following aspects:(1)A switch panel defect dataset was created: A total of 2346 switch panel surface images were collected in this dataset,and defect labeling was performed on the captured switch panel surface image samples using the Label Img image annotation tool.(2)The small sample problem and small target problem in the panel surface defect detection task were studied,and two defect detection algorithms based on multitask learning and CNN Transformer hybrid model were proposed.The experimental demonstration was carried out on the Kolektor SDD metal panel data data set with a small number of defect images and the PCB panel data data set with a small number of defect targets,respectively.The results showed that the two defect detection algorithms proposed in this article have achieved detection accuracy of over 98% on their corresponding experimental datasets.(3)The defect detection algorithm based on the CNN-Transformer hybrid model proposed in this article is applied to the switch panel surface defect detection task,and based on this,a switch panel defect detection software system with a GUI interface is designed to further verify whether the algorithm proposed in this article has good versatility and whether the algorithm can also achieve good defect detection results in practical application scenarios.