Vision-based automated optical inspection for real-time quality control: hybrid single-camera deep-learning AOI for mixed-SKU bottling lines

Household-chemical bottling lines require high-speed verification of closure integrity, label conformity, expiry-code legibility, and container deformation, yet many existing solutions are either cost-prohibitive multi-camera automated optical inspection (AOI) tunnels or unreliable manual inspection. This paper presents a low-cost single-camera inspection station for mixed stock-keeping-unit (SKU) production that integrates deep-learning-based localization with deterministic, tolerance-based AOI. A fixed USB camera ( 1280 × 720 px) with ring-light illumination acquires images downstream of the expiry printer at a line-synchronous rate. YOLOv8 performs SKU classification and region selection, while task-specific modules assess cap sealing via hybrid fusion (height-profile AOI with a learned backup classifier), label/color conformity, volume–SKU consistency, expiry-date optical character recognition (OCR), and structural deformation. Per-bottle OK/NG decisions and module-level scores are logged to support traceability. On 33 mixed-SKU bottles, hybrid fusion achieved perfect cap-sealing classification (vs. 90.9% AOI-only and 87.9% YOLO-only), color/SKU accuracy of 100%, deformation detection of 97.0% (100% sensitivity, 94.4% specificity), expiry-date OCR accuracy of 87.9%, and label-conformity accuracy of 90.9%. These results show that, under a calibrated low-cost optical setup, commodity hardware can deliver practical multi-defect inline inspection when learning-based region selection, deterministic AOI rules, and traceability are co-designed.