摘要
针对环锭纺纱过程中断纱检测效率低下、易受环境干扰等问题,提出一种改进的MobileNetV3-Small断纱检测方法。首先,设计并构建基于四轮四驱移动机器人平台的图像采集系统,获取7个类别断纱数据集。其次,采用Prewitt算子和伽马校正进行图像增强,提升纤维特征的清晰度。然后,在MobileNetV3-Small模型中引入改进后的CoordAttention注意力机制,通过增加通道注意力机制,增强模型对纱线空间和通道信息的捕捉能力。最后,为解决类别不平衡和特征学习问题,采用Focal Loss和Center Loss的联合损失函数,提升模型的泛化能力和分类精度。试验结果显示:模型在验证集上取得了97.8%的准确率,相比基准模型提升了3.2个百分点,模型参数量降低至4.42 M,体现了其轻量化和高效性的优势。该研究提出的模型在断纱检测任务中具有较高的准确性和实时性,能够有效解决传统方法的不足。
Aiming the issues of lower test efficiency in yarn breakage detection and environmental susceptibility during ring spinning process,a modified MobileNetV3-Small yarn breakage detection method was proposed.Firstly,an image acquisition system based on four-wheel-four-drive mobile robot platform was designed and constructed,then seven category yarn breakage dataset was obtained.Secondly,Prewitt operator and gamma correction were used for image enhancement,the clarity of fiber features was improved.Next,modified CoordAttention mechanism was integrated in MobileNetV3-Small model.Capture ability of the model to yarn room and channel information was enhanced through adding channel attention mechanism.Finally,to solve the issues of class imbalance and feature learning,combined loss function of Focal Loss and Center Loss was adopted to enhance the model's generalization ability and classification precision.The experimental results showed that accuracy of the model was 97.8% on the validation set.Compared to the baseline model,the accuracy was increased 3.2 percentage points,the number of model parameters was decreased 4.42 M,which highlighting its advantages in lightweight design and high efficiency.The accuracy and real-time performance of yarn breakage detection tasks in the proposed model were higher,which could effectively overcome the limitation of the traditional methods.
作者
付奇强
王升
国冰磊
程凯
黄兴宇
陈佳
FU Qiqiang;WANG Sheng;GUO Binglei;CHENG Kai;HUANG Xingyu;CHEN Jia(College of Computer and Cyber Security,Fujian Normal University,Fuzhou,350117,China;Quanzhou Equipment Manufacturing Research Center,Haixi Institute of Chinese Academy of Sciences,Fuzhou,362200,China;Computer Engineering School,HuBei University of Arts and Sciences,Xiangyang,441053,China)
出处
《棉纺织技术》
2025年第3期50-58,共9页
Cotton Textile Technology
基金
湖北省自然科学基金项目(2022CFB805)
福建省科技计划项目(2023T3086)。
