一种改进Mask R-CNN的化妆棉棉片缺陷检测方法  

An improved Mask R-CNN defect detection method for cotton pads

作　　者：李亮[1] 陈广锋[1] 丁彩红[1] LI Liang;CHEN Guangfeng;DING Caihong(College of Mechanical Engineering,Donghua University,Shanghai 201620,China)

机构地区：[1]东华大学机械工程学院,上海201620

出　　处：《东华大学学报（自然科学版）》2023年第5期78-87,共10页Journal of Donghua University(Natural Science)

摘　　要：针对化妆棉棉片缺陷人工检测效率低、精度低的问题,提出一种改进Mask R-CNN的化妆棉棉片缺陷检测方法。在Mask R-CNN的基础上使用ResNet50作为特征提取网络,引入深度卷积网络来提高缺陷特征的学习能力。通过设计多信息融合特征金字塔网络来提高小面积缺陷的检测,引入注意力机制模块来减少漏检和误检现象,构造优化的损失函数来降低样本不平衡对结果的影响。通过试验验证了该算法的有效性,结果表明,改进后的Mask R-CNN模型平均检测精度达95.7%,召回率达88.1%,整体性能明显优于原始的Mask R-CNN、Faster R-CNN、SSD和YOLOv5算法模型,能准确检测出常见的化妆棉棉片缺陷。Aiming at the problems of low efficiency and precision in manual detection of cotton pad defects,an improved Mask R-CNN defect detection method is proposed.On the basis of Mask R-CNN,ResNet50 is used as the feature extraction network,and depth convolution is introduced to improve the learning ability of defect features.The multi-information fusion feature pyramid network is designed to improve the detection of small defects,an attention mechanism module is introduced to reduce missed and false detections,and the optimized loss function is constructed to reduce the impact of sample imbalance.The effectiveness of the algorithm was verified by experiments.The results show that after the average detection accuracy of the optimized Mask R-CNN algorithm model reaches 95.7%and the recall rate reaches 88.1%.The overall performance of the model is significantly better than the original Mask R-CNN,Faster R-CNN,SSD and YOLOv5 algorithm models.The optimized Mask R-CNN model can accurately detect common defects of cotton pads,which have a good effect on the detection of cotton pads defects.

关 键 词：化妆棉棉片 缺陷检测 Mask R-CNN 深度卷积 特征融合 注意力机制 

分 类 号：TS959.9[轻工技术与工程] TP391.4[自动化与计算机技术—计算机应用技术]