<![CDATA[Packaging quality is an important factor in ensuring product quality and safety, particularly for can packaging that is widely used in food and beverage industry. During the production process, defects such as dents, scratches, and label damage may occur on both the inner and outer surfaces of cans, potentially posing health risks to consumers. Therefore, an automated inspection system capable of performing comprehensive, accurate, and rapid inspection is required. This study proposes a can packaging defect inspection system based on the single-stage object detection algorithm YOLOv8n with a multi-view frame fusion approach in a manufacturing system. The multi-view frame fusion approach utilized two cameras positioned in opposite directions and sampling frames from video data to generate a multi-view visual representation without increasing system complexity. The dataset was constructed by merging multi-view images of normal and defective cans and was then used to train the YOLOv8n model. Experimental results show that the proposed system achieved a precision of 0.982, a recall of 0.909, a mean Average Precision at IoU 0.5 of 0.962, and an inference speed of up to 29 frame per second. The direct testing results demonstrate that the system could operate reliably with a low detection error rate. These results indicate that the proposed multi-view frame fusion approach combined with YOLOv8n is effective and feasible for implementation as an automated quality control system in manufacturing environments.]]>
Copyrights © 2026
