摘要
为提高织物瑕疵图像分类准确率,提出了一种基于广义奇异值分解的加权线性判别分析(Linear discriminant analysis, LDA)织物瑕疵图像分类算法。首先,使用加权均值保留瑕疵特征信息,反映真实数据的分布情况;然后,对类间散射矩阵和类内散射矩阵分别添加类间离散权重和类内紧凑权重,克服了其他织物瑕疵分类算法忽视局部几何信息的问题;最后,结合广义奇异值分解,解决了小样本奇异不可逆的问题,提升了计算效率。不同织物数据集实验结果表明,该算法能有效解决LDA存在缺乏局部几何信息和小样本高维的问题;同时相较于其他基于LDA的分类算法,能取得更好的分类准确率,且分类所需的计算时间具有一定竞争力。
To improve the accuracy of fabric defect image classification, a weighted linear discriminant analysis(LDA) algorithm based on generalized singular value decomposition was proposed for fabric defect classification. First, the defect feature information was retained by using the weighted means to reflect the distribution of real data. Then, the inter-class separation weight and intra-class aggregation weight were added into the inter-class scattering matrix and the intra-class scattering matrix, respectively, which tackled with the neglection of local geometric information occurring in other fabric defect classification algorithms. Finally, by combining the generalized singular value decomposition method, the problem of singular irreversibility for small samples was solved and the computational efficiency was improved. The experimental results based on different fabric image datasets demonstrate that the proposed method can effectively solve the lack of local geometric information features and the high-dimensionality of small samples. At the same time, compared with other state-of-the-art LDA-based algorithms, the proposed method has achieved better classification accuracy, and the computational time required for classification is more competitive.
作者
钟佳莹
吕文涛
陈亮亮
王成群
ZHONG Jiaying;LU Wentao;CHEN Liangliang;WANG Chengqun(School of Information Science and Technology,Zhejiang Sci-Tech University,Hangzhou 310018,China;School of Applied Engineering,Zhejiang Institute of Economics and Trade,Hangzhou 310018,China)
出处
《浙江理工大学学报（自然科学版）》
2022年第3期313-322,共10页
Journal of Zhejiang Sci-Tech University(Natural Sciences)
基金
国家自然科学基金项目(61601410)
浙江省科技厅重点研发计划项目(2021C01047)
东北大学流程工业综合自动化国家重点实验室联合基金(2021-KF-21-03,2021-KF-21-06)。
关键词
线性判别分析
图像分类
广义奇异值分解
局部几何信息
小样本
linear discriminant analysis
image classification
generalized singular value decomposition
local geometric information
small samples
