【摘要】 为提升冰箱柜门表面缺陷的检测效率与准确率，提出一种基于YOLO系列模型的快速检测方法。研究构建了包含划痕、凹陷和正常表面三类样本的高质量工业图像数据集，结合多角度采集与多种数据增强策略，有效增强了模型对复杂表面反射与多尺度缺陷的适应能力。选取YOLOv5、YOLOv8、YOLOv10与YOLO11四种主流目标检测模型开展对比实验，分别从检测精度、召回率、推理速度和模型规模等方面进行评估。实验结果表明，YOLOv8n模型在准确率(AP=88.0%)、召回率(92.0%)和推理效率方面表现最优，最适合作为冰箱生产线的实时检测模型; YOLO11n模型具备较高的召回能力和轻量化特性，更适合边缘计算场景下的精密质检任务。该研究为冰箱表面缺陷智能检测模型的选型与部署提供了技术依据。更多还原

【Abstract】 To improve the efficiency and accuracy of refrigerator door surface defect detection,a rapid detection approach based on the YOLO model series was investigated. A high-quality industrial dataset containing scratches,dents,and normal surface samples was constructed,incorporating multi-angle image acquisition and diverse data augmentation strategies to enhance model adaptability to complex surface reflections and multi-scale defects. Four mainstream object detection models—YOLOv5,YOLOv8,YOLOv10,and YOLO11—were evaluated through comparative experiments focusing on detection accuracy,recall,inference speed,and model size. The results demonstrated that YOLOv8n achieved the highest performance in terms of accuracy(AP = 88. 0%), recall(92. 0%), and inference efficiency,making it suitable for real-time production line deployment. YOLO11n exhibited strong recall and lightweight architecture,supporting its application in edge-computing-based precision inspection. This study provides a technical reference for selecting and deploying intelligent surface defect detection models in refrigerator manufacturing.更多还原