AI-Based Automated Visual Inspection for Multi-Defect Detection in Footwear Manufacturing

Most footwear factories still check shoes by eye. This is slow, and different inspectors often reach different decisions. This paper presents DefectGuard, a camera-based system that finds and classifies visible defects in shoes and gives each shoe a PASS, FAIL or REVIEW result. The system uses a YOLOv8m object detector trained on 6,240 images of 14 shoe models, covering eight defect types such as scratches, stitching faults and excess glue. On the test set, DefectGuard reached a precision of 0.908, a recall of 0.878 and an mAP@0.5 of 0.910, which is better than the three baseline models compared. It processes one image in 18.6 ms on an NVIDIA RTX 4090 GPU. Using a higher input resolution raised recall on small defects from 0.71 to 0.79. In a six-week pilot on a production line, the system inspected 38,400 shoes and its analytics helped the factory find and fix a recurring glue problem. These results show that AI-based inspection is practical for everyday footwear quality control.