Our solution detects, classifies, and counts cigarette cartons and boxes discharged from an Automatic Dispensing System (APS) as they move along a high-speed conveyor at 1.5 m/s. Powered by AiVOps, our MLOps platform, the system enables rapid on-site labeling, training, and model deployment, allowing seamless adaptation to newly designed or previously unseen products. 

To optimize box recognition, we have designed and developed a specialized mechanical structure and optical system. The solution employs a two-step deep learning approach: first, it accurately identifies the position of the boxes, and then it precisely counts them, ensuring reliable performance even in complex environments. 

By incorporating data augmentation and class imbalance correction, the system enhances robustness across diverse scenarios. When needed, AiVOps facilitates additional training, ensuring flexibility and stable operation even when encountering new product variations or challenging edge cases.