Surface Defect Detection and Classification Based on Fusing Multiple Computer Vision Techniques

Computer vision techniques are widely used for automated quality control in production line, which can identify defects in from collected images. Due to unclear features, diverse product shapes, and small defect sample size, a single computer vision method can hardly achieve the task of product surface defect detection with high accuracy and high efficiency. Thus, we proposed a novel approach based on fusing multiple computer vision methods, and combining online models with offline models. The proposed approach can achieve high detection accuracy in real-time over the whole production process. We also implemented the system and obtained excellent results in a case study of surface defect detection of lipstick. This research shows that new information and intelligence technologies have their importance in the quality control of manufacturing.