改进YOLOv8的织物疵点检测算法  

Improved fabric defect detection algorithm of YOLOv8

作　　者：丁琼 祝双武 田乐 王茹 余灵婕 Ding Qiong;Zhu Shuangwu;Tian Le;Wang Ru;Yu Lingjie(College of Textile Science and Engineering,Xi’an Polytechnic University,Xi’an 710048,China)

机构地区：[1]西安工程大学纺织科学与工程学院,西安710048

出　　处：《现代计算机》2024年第5期24-30,共7页Modern Computer

基　　金：陕西省教育厅科研计划项目(18JS042);中国纺织工业联合会科技指导性项目(2019057)。

摘　　要：针对传统织物疵点检测方法耗时耗力,主流检测模型速度慢或精度低的问题,改进了YOLOv8模型,旨在提高织物疵点检测的性能。选择性注意力模块(LSKBlock)融入YOLOv8主干网络,使模型自动学习并决定哪些信息值得关注与处理,优化模型对相关信息的处理效果;在颈部将卷积层替换为可变形卷积(DCN),加强网络感知目标形变的能力,进一步提升网络的特征提取和定位能力;另外通过设计一种轻量型范式(Slimneck),提高模型精度并降低复杂度。在TILDA和织物疵点数据集上进行性能评估,结果显示,改进的YOLOv8模型在两个数据集上的mAP分别达到88.6%和92.7%,相较原模型提高了4.1和4.0个百分点,且检测速度满足实际生产中的要求。Aiming at the problems of time-consuming and labor-intensive traditional fabric defect detection methods and slow speed or low precision of mainstream detection models,YOLOv8 model was improved to improve the performance of fabric defect detection.The Selective Attention module(LSKBlock)is integrated into the YOLOv8 backbone network,which enables the model to automatically learn and decide which information deserves attention and processing,and optimize the model’s processing effect on relevant information.The convolutional layer is replaced by deforming convolution(DCN)at the neck to enhance the ability of the network to perceive the deformation of the target,and further improve the feature extraction and localization capability of the network.In addition,a lightweight paradigm(Slimneck)is designed to improve the accuracy and reduce the complexity of the model.Performance evaluation was performed on TILDA and fabric defects datasets.The results showed that the mAP of the improved YLOLv8 model on the two datasets reached 88.6%and 92.7%respectively,4.1 and 4.0 percentage points higher than that of the original model,and the detection speed met the requirements in actual production.

关 键 词：织物疵点检测 YOLOv8 选择性注意力 轻量化 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TP391.41[自动化与计算机技术—控制科学与工程] TS107[轻工技术与工程—纺织工程]