摘要
针对目前织物自动开幅设备无法准确识别复杂织物纹理背景下的开幅引导线的问题,设计了一种基于迁移学习的GoogleNet网络与非迁移学习的残差网络的织物纹理特性分类系统。所使用的样本数据集分为密集型花纹等7种纹理特性,共计1543张图,随机选取80%的图片作为训练集,剩余20%图片作为测试集。两种不同的卷积神经网络在现有的数据集中达到了100%的识别准确率,并且在后续的系统测试中,新增了300张样品,最终系统的识别准确率达到了98%。实验结果表明,将GoogleNet网络与残差网络应用于织物纹理特性的分析与分类切实可行,以此为算法基础构建的系统具有实用价值。
Aiming at the problem that the current fabric automatic opener can’t accurately identify the opener guide line under the complex fabric texture background,this paper designs a fabric texture feature classification system based on GoogleNet network of transfer learning and residual network of nontransfer learning.The sample data set is divided into seven texture characteristics,such as dense pattern,a total of 1543 images,randomly selected 80%of the images as the training set,the remaining 20%of the images as the test set.Two different convolutional neural networks achieve 100%recognition accuracy in the existing data set,and in the subsequent system test,300 samples are added,and the final recognition accuracy of the system reaches 98%.The experimental results show that the application of GoogleNet network and residual network to the analysis and classification of fabric texture characteristics is feasible,and the system based on this algorithm has practical value.
作者
邓宇平
王桂棠
Deng Yuping;Wang Guitang(School of Electromechanical Engineering,Guangdong University of Technology,Guangzhou 510006,China)
出处
《电子测量技术》
北大核心
2021年第7期31-38,共8页
Electronic Measurement Technology
