Sewing defect detection using YOLOv8: a deep learning approach

Purpose The primary objective of this study was to evaluate the performance of the YOLOv8 ((You Only Look Once, version 8)) model in detecting several classes of sewing defects, such as skipped stitches, overlapped stitches, stains and damage and bobbin thread pull-up. The purpose of this study was to develop an effective and accurate method for detecting sewing defects. Design/methodology/approach Initially, data pre-processing and image annotation were performed on the sewing defects, after which YOLOv8 was trained on the annotated data set. The performance was measured using parameters such as precision, recall, mean Average Precision (mAP) and inference time. Findings The YOLOv8 model performed well with a high detection accuracy of 90.5% mAP@0.5 and an average inference time of 24 ms. Practical implications The results indicate that YOLOv8 is efficient in detecting sewing defects in garments, where minor defects can considerably affect garment quality. Originality/value Although prior research has investigated YOLO-based approaches, a notable gap persists in their application to the detection of multi-class sewing defects. To address this, we employ the YOLOv8 model to classify various types of sewing defects.