西安工程大学学报2025，Vol.39Issue(5)：100-108,9.DOI:10.13338/j.issn.1674-649x.2025.05.012
一种改进YOLOv8的陶瓷环缺陷检测算法
An improved YOLOv8 ceramic ring defect detection algorithm
摘要
Abstract
To achieve efficient and high-precision defect detection of ceramic rings,this paper im-proves YOLOv8 and proposes a new ceramic ring defect detection algorithm,namely YOLOv8-SDE algorithm.Firstly,in order to improve computational efficiency and increase the algorithm's attention to ceramic ring defect features,a lightweight module StarNet Block and an efficient multi-scale attention(EMA)mechanism were used to design a CSTEA module to replace the first two layers of C2f modules in the backbone network.Secondly,to solve the problem of difficulty in extracting defect features,a deformable convolution was used to replace the traditional convolu-tion in the residual structure of the last two layers of C2f in the backbone network,forming the CDC3 module.Finally,to further improve the lightweight level of the algorithm and its perception of defect features,a lightweight neck network was used to replace the original neck network,and experiments were conducted on a self-made dataset.The results showed that the improved algo-rithm significantly reduced the computational complexity,parameter complexity,and weight file volume.The mean average precision(mAP)increased by 2.8 percentage points,and the mAP value for various defects in ceramic rings reached 92.3%,at the same time,it required less com-putational resources while maintaining higher detection accuracy.关键词
缺陷检测/轻量化网络/StarNet/陶瓷零件/高效多尺度注意力(EMA)机制Key words
defect detection/lightweight network/StarNet/ceramic components/efficient multi-scale attention(EMA)mechanism分类
信息技术与安全科学引用本文复制引用
管声启,杨振,党慧,宋翌宸..一种改进YOLOv8的陶瓷环缺陷检测算法[J].西安工程大学学报,2025,39(5):100-108,9.基金项目
陕西省重点研发计划项目(2022GY-058) （2022GY-058）
西安市创新能力强基计划-人工智能技术攻关项目(21RGZN0021) （21RGZN0021）
