| Car painting is an important process in automobile manufacturing.Painting defects on the car surface need to be checked and fixed after painting.Defect detection system in traditional industrial is based on human vision checking and verifying.Human checking is of low efficiency due to limited resolution and speed of human vision,individual bias,and visual fatigue of workers caused by long intensive work and reflection of white light.This paper states importance of car painting quality,summarizes current research about car painting defect detection,analyzes pros and cons between traditional object detection algorithms and those based on deep learning,and proposed a car body paint defect detection and classification method,which overcomes disadvantages of traditional human vision detecting system and improve the quality of car painting.This paper mainly includes following contents:(1)The dataset of car paint defect is acquired by collecting data in car painting workshop.By analyzing types and characteristics of the dataset,the author proposed an offline data augmentation strategy to enlarge the data and build a database of car paint defect.(2)The author proposed an improved Mobile Net-SSD algorithm by altering network structure and matching strategy based on SSD algorithm.(3)Design and implement an automated car paint defect detection and classification system.The system provides users with detection and classification service by Web server.The system is compatible with different operating systems and devices and ensures consistent user experience.Besides,the system provides HTTP interfaces for developing and integration into other systems if necessary.The experiment results show that the improved Mobile Net-SSD algorithm proposed has significantly better performance compared with SSD algorithm both in accuracy and speed.The detection and classification system implemented effectively improves deficiencies of traditional human vision detection method. |
PDF Full Text Request