棉纺织技术2026，Vol.54Issue(7)：57-62,6.DOI:10.26967/j.issn1000-7415.202503010
基于改进ResNet18模型的异性纤维识别分类方法研究
Research on foreign fiber recognition and classification method based on improved ResNet18 model
摘要
Abstract
Aiming at the problems of lower efficiency and insufficient accuracy for manual detecting foreign fibers in cotton,a deep learning method based on improved ResNet18 was proposed to realize automatic and accurate recognization of foreign fibers,improve the quality control level of textiles.Combined with rotation,scale,shift and other foreign fiber image fusion,robustness of the improved ResNet18 model to complex scenes was improved.By adding 1×1 convolutional neural network to skip connection channels,the problem of deep network gradient disappearance was solved,feature fusion ability of the improved ResNet18 model was enhanced.Adam optimizer was used to dynamically adjust the learning rate and accelerate convergence of the improved ResNet18 model,so that the accuracy rate on foreign fiber test set was reached 99.9%,which was 4.8 percentage points higher than the original ResNet18 model.The accuracy and efficiency of foreign fiber detection was increased by improved ResNet18 model effectively through structural optimization and data enhancement,which provided an intelligent foreign fiber detection scheme for textile industry.关键词
ResNet18/异性纤维/纤维识别/深度学习/图像分类Key words
ResNet18/foreign fiber/fiber recognition/deep learning/image classification分类
轻工纺织引用本文复制引用
于春明,张洪卫,塔依尔·亚森,图尔贡江·乌热依木,朱若斐,宋均燕,王丹..基于改进ResNet18模型的异性纤维识别分类方法研究[J].棉纺织技术,2026,54(7):57-62,6.基金项目
新疆维吾尔自治区自然科学基金青年基金项目(2024D01C213,2024D01C214) （2024D01C213,2024D01C214）
新疆大学博士启动基金项目(620323015) （620323015）
新疆东纯兴纺织有限公司横向项目(202410140007) （202410140007）
