摘要
为解决羊绒羊毛纤维图像中类间差异小、类内差异大导致难以准确分类的问题,提出一种融合特征选择与聚类优化的细粒度羊绒羊毛纤维分类方法。提取羊绒羊毛的形态、纹理和关键点特征,来表征羊绒羊毛纤维的细微差异;采用基于类内类间距离的特征选择方法,筛选出判别性强的特征减少冗余信息;利用类内马氏距离与类间欧式距离调整策略优化K-means聚类,识别并处理离群点,减少其对簇中心的干扰,增强簇内样本的紧密性,提升类间可分性。结果显示:该方法在羊绒羊毛纤维图像数据集上的分类准确率达到98.92%,相比基线模型提升了3.04%。同时,类内离散度整体下降约37%,类间分离度提升约26%,表明所提方法在细粒度纤维分类中具有显著优势。
To address the challenge of low inter-class variance and high intra-class variance in cashmere and wool fiber image classification,this paper proposed a fine-grained classification method that integrates feature selection and clustering optimization.First,morphological,textural,and keypoint features of cashmere and wool fibers were extracted to capture subtle differences among fiber types.Then,a feature selection method based on intra-class and inter-class distance was employed to identify highly discriminative features and reduce redundant information.Finally,an improved K-means clustering algorithm was introduced,incorporating Mahalanobis distance for intra-class similarity and Euclidean distance for inter-class separation to identify and mitigate the influence of outliers.This enhances intra-cluster compactness and inter-cluster separability.The experimental results demonstrate that the proposed method achieves a classification accuracy of 98.92%on the cashmere/wool fiber image dataset,representing a 3.04%improvement over the baseline model.Simultaneously,the intra-class dispersion decreases by approximately 37%overall,while the inter-class separation increases by about 26%,indicating that the proposed method exhibits significant advantages in fine-grained fiber classification.
作者
顾梅花
候嘉乐
朱耀麟
韩李婷
GU Meihua;HOU Jiale;ZHU Yaolin;HAN Liting(School of Electronics and Information,Xi’an Polytechnic University,Xi’an 710048,China)
出处
《纺织高校基础科学学报》
2025年第5期88-97,共10页
Basic Sciences Journal of Textile Universities
基金
陕西省科技厅面上项目(2024JC-YBMS-491)
陕西省科技厅自然科学基金重点项目(2023-JC-ZD-33)
西安市科技局重点产业链技术攻关项目(23ZDCYJSGG0008-2023)。
关键词
细粒度纤维图像分类
类内马氏距离
类间欧式距离
特征选择
改进K-MEANS
fine-grained fiber image classification
intra-class mahalanobis distance
inter-class euclidean distance
feature selection
improved K-means
