Surface Defects Detection of Car Door Seals Based on Improved YOLO V3

Abstract

At present, the surface defects of car door seals are generally detected by manual operation, which is inefficient and costly. This paper presents a new detection method to find the surface defects of car door seals with the application of modified YOLO V3 algorithm. Firstly, the K-means clustering algorithm was employed to analyze the number and size of the candidate box of the target detection. Then, the feature pyramid was constructed in the first module of the trunk feature extraction network. Finally, the attention mechanism SE module was added in the output part of darknet53 and two feature gold towers extraction network of the trunk feature. The improved YOLO V3 algorithm was carried out on the defect data set of automobile door seals. The results indicate that compared with the original YOLO V3 algorithm, the improved YOLO V3 mAP reaches 52.71%, increasing by 3.28%.