无线电工程2025，Vol.55Issue(7)：1440-1447,8.DOI:10.3969/j.issn.1003-3106.2025.07.009
基于改进YOLOv10的白酒瓶盖缺陷检测算法
Improved YOLOv10-based Defect Detection Algorithm for Liquor Bottle Caps
摘要
Abstract
In view of the problem of cap defects in the liquor production process,YOLOv10-the latest product of the YOLO series-is outstanding due to its high efficiency and accuracy.However,when applied to cap defect detection,it still has limitations due to challenges such as variable defect sizes,abundant local features,and unbalanced samples.Therefore,an improved liquor cap defect detection algorithm based on improved YOLOv10.Firstly,the Slide Loss function is introduced to eliminate the imbalance of positive and negative samples in the cap dataset.Secondly,the deformable convolution enhancement C2f module is used to mitigate the influence of diverse defect geometries on the convolution.In addition,the Large-Separable-Kernel-Attention(LSKA)module is introduced to improve the computational efficiency and memory utilization of the model.Then,the Partial Self-Attention(PSA)mechanism of the backbone network is replaced by the Convolutional Block Attention Module(CBAM),enhancing the model's ability to capture local features.Finally,the Content-Aware ReAssembly of Features(CARAFE)upsampling module is introduced to improve the accuracy.Experimental results show that compared with the original YOLOv10 model,the floating-point operations of the proposed model is reduced by 9％,and the average accuracy is increased by 3.1％to 76.8％,which effectively improves the accuracy of cap defect detection.关键词
缺陷检测/白酒瓶盖/YOLOv10/损失函数/可变形卷积/注意力机制/上采样算子Key words
defect detection/liquor bottle cap/YOLOv10/loss function/deformable convolution/attention mechanism/upsampling operator分类
信息技术与安全科学引用本文复制引用
吴佳坤,卿粼波,谢建斌,郑建国,李益..基于改进YOLOv10的白酒瓶盖缺陷检测算法[J].无线电工程,2025,55(7):1440-1447,8.基金项目
四川大学-遂宁校市合作专项项目(2022CDSN-11)Special Project for Sichuan University-Suining City Cooperation(2022CDSN-11) （2022CDSN-11）
