摘要
针对传统人工检测方法对石英坩埚质检主观性强、误检率高,文章提出一种基于YOLOv11n的轻量化缺陷检测模型MCS-YOLO。首先,在YOLOv11n基础上采用多尺度卷积模块设计全新的C3k2结构,以提升对微小缺陷的感知精度;其次,在颈部网络中嵌入卷积块注意力模块,引导网络聚焦于关键缺陷区域;最后,采用SCYLLA-IoU(SIoU)边界框损失函数优化回归路径,提升目标定位性能。实验结果表明,该算法在自建石英坩埚数据集上平均精度均值达到92.7%,较基准模型提升3.4%,参数量为3.02 M,计算量为7.26 GFLOPs,权重仅为6.30 MB,兼具高精度、轻量化与高鲁棒性,适合应用于工业检测场景。
To address the problems of strong subjectivity and high false positive rate in the traditional manual inspection method for quartz crucible quality inspection,a lightweight defect detection model MCS-YOLO based on YOLOv11n is proposed.First,the C3k2 structure is designed on the basis of YOLOv11n,which adopts the multi-scale convolution module(MSCB)to improve the perception accuracy of tiny defects.Second,the convolutional block attention module(CBAM)is embedded into the neck network to guide the network to focus on key defect areas.Finally,the SCYLLA-intersection over union(SIoU)bounding box loss function is used to optimize the regression path and improve target positioning performance.Experimental results show that the algorithm achieves a mAP@0.5 of 92.7% on the self-built quartz crucible dataset,which is 3.4% higher than that of the baseline model.With 3.02 M parameters requiring 7.26 GFLOPs of calculation and a weight of only 6.30 MB,the model features high precision,lightweight performance,and strong robustness,making it suitable for applications in industrial detection scenarios.
作者
黄晶晶
宋宇辉
赵谦
HUANG Jingjing;SONG Yuhui;ZHAO Qian(Jiangsu Shuangliang Low-Carbon Industrial Technology Research Institute Co.,Ltd.,Wuxi 214400,CHN;School of Communication and Information Engineering,Xi’an University of Science and Technology,Xi’an 710000,CHN;Xi’an Dishan Vision Technology Limited Company,Xi’an 712044,CHN)
出处
《半导体光电》
北大核心
2025年第6期1079-1087,共9页
Semiconductor Optoelectronics
基金
陕西省科技成果转化计划-百项科技成果转化行动项目(2023-YD-CGZH-29)
陕西秦创原“科学家+工程师”队伍建设项目(S2023-YF-LLRH-QCYK-0271)
西安市科技计划(科学家+工程师队伍建设项目)(24KGDW0030).
