摘要
为降低厂家因瓶装酒瑕疵带来的不必要损失,提出一种改进的Cascade R-CNN算法模型,对酒瓶瑕疵进行检测。采用基于聚类算法的Anchor生成策略,将多尺度预测的骨干网络用作特征提取,使用感兴趣对齐层取代原先的感兴趣池化层。将改进的模型与其它基于Faster R-CNN和Cascade R-CNN的酒瓶瑕疵检测模型做消融实验,实验结果表明,该模型能够更加准确识别和定位出多类酒瓶瑕疵情况。在检测速度方面虽然略逊于其它模型,但模型检测的准确度达到了79.6%,远高于其它模型。
To reduce the unnecessary losses caused by bottled wine defects,an improved Cascade R-CNN algorithm model was proposed to detect bottle defects.The Anchor generation strategy based on clustering algorithm was adopted,the backbone network of multi-scale prediction was used as feature extraction,and the alignment layer of interest was used to replace the original pooling layer of interest.The improved model was used in ablation experiments with other bottle defect detection models based on Faster R-CNN and Cascade R-CNN.Experimental results show that the model can more accurately identify and locate multiple types of bottle defects.Although it is slightly inferior to other models in terms of detection speed,the accuracy of model detection reaches 79.6%,which is much higher than other models.
作者
高林
张玉蓉
李升凯
朱庆港
姜旭辉
GAO Lin;ZHANG Yu-rong;LI Sheng-kai;ZHU Qing-gang;JIANG Xu-hui(School of Automation and Electronical Engineering,Qingdao University of Science and Technology,Qingdao 266100,China)
出处
《计算机工程与设计》
北大核心
2022年第2期434-442,共9页
Computer Engineering and Design
基金
山东省自然科学基金项目(ZR2014FL018)
青岛科技大学博士科研基金项目(010022530)
教育部产学合作协同育人基金项目(施耐德公司,2018第一批)。
