基于改进YOLOv8n的织物疵点检测算法  

Fabric defect detection algorithm based on improved YOLOv8n

作　　者：陈泽纯 林富生 张庆 宋志峰 刘泠杉 余联庆 CHEN Zechun;LIN Fusheng;ZHANG Qing;SONG Zhifeng;LIU Lingshan;YU Lianqing(College of Mechanical Engineering and Automation,Wuhan Textile University,Wuhan,Hubei 430200,China;Three Dimensional Textile Engineering Research Center of Hubei,Wuhan,Hubei 430200,China;Hubei Key Laboratory of Digital Textile Equipment,Wuhan,Hubei 430200,China;College of Textile Science and Engineering,Wuhan Textile University,Wuhan,Hubei 430200,China)

机构地区：[1]武汉纺织大学机械工程与自动化学院,湖北武汉430200 [2]三维纺织湖北省工程研究中心,湖北武汉430200 [3]湖北省数字化纺织装备重点实验室,湖北武汉430200 [4]武汉纺织大学纺织科学与工程学院,湖北武汉430200

出　　处：《毛纺科技》2025年第2期118-126,共9页Wool Textile Journal

基　　金：国家留学基金管理委员会资助项目(202310810005)。

摘　　要：针对织物疵点检测算法计算复杂且难以部署,织物缺陷目标小、尺度变化大导致检测精度低、易漏检等问题,提出基于改进YOLOv8n的织物疵点检测算法(YOLOv8n-CMB)。首先在主干网络引入轻量级通道优先卷积注意力模块(CPCA),实现在通道和空间维度上注意力权重的动态分配,提高网络对小目标的感知能力;其次将Backbone层中的部分C2f模块替换为轻量级MobileViTv3模块,以扩大网络感受野,增强模型对小目标缺陷的检测能力;最后使用双向特征金字塔网络(BiFPN)改进路径聚合网络(PANet),有效解决网络冗余和计算量复杂的问题。实验结果表明:YOLOv8n-CMB算法与原始YOLOv8n模型相比,平均精度均值提高了4.4个百分点,参数量减少了10%,在保证模型轻量化的同时具有较高的检测精度,为纺织领域提供了更好的检测思路。Aiming at the problems such as complex calculation and difficult deployment of fabric defect detection algorithm,low detection accuracy and easy omission due to small defect target and large scale change,a fabric defect detection algorithm based on improved YOLOv8n(YOLOv8n-CMB)was proposed.Firstly,the lightweight channel priority Convolutional attention module(CPCA)was introduced into the backbone network to realize the dynamic allocation of attention weights in the channel and spatial dimensions,and the network′s awareness of small targets was improved.Secondly,some C2f modules in Backbone layer were replaced with lightweight MobileViTv3 modules to expand the network receptor field and enhance the model′s detection ability of small target defects.Finally,the bidirectional feature pyramid network(BiFPN)was used to improve the path aggregation network(PANet)to solve the problem of network redundancy and complex computation.The test results show that compared with the original YOLOv8n model,the average accuracy of the YOLOv8n-CMB algorithm was increased by 4.4 percentage points and the number of parameters was reduced by 10%.It has high detection accuracy while ensuring the lightweight of the model,and provides a better detection idea for the textile field.

关 键 词：YOLOv8n MobileViTv3 特征金字塔 注意力模块 织物疵点 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]