融合空间信息的词袋模型用于羊绒羊毛纤维识别  

Identification of cashmere and wool fiber based on bag of words model with spatial information

作　　者：朱耀麟[1,2] 穆婉婉 武桐[1] ZHU Yaolin;MU Wanwan;WU Tong(College of Electronics and Information,Xi′an Polytechnic University,Xi′an,Shaanxi 710048,China;College of Electronics and Information,Northwestern Polytechnical University,Xi′an,Shaanxi 710072,China)

机构地区：[1]西安工程大学电子信息学院,陕西西安710048 [2]西北工业大学电子信息学院,陕西西安710072

出　　处：《毛纺科技》2021年第6期75-79,共5页Wool Textile Journal

基　　金：陕西省科技厅重点研发计划一般项目(2019GY-098);陕西省教育厅服务地方科学研究计划项目(18JC012);陕西省教育厅重点研究计划产业用纺织品协同创新中心项目(20JY026);榆林市科技局科创新城项目(2018-2-24);绍兴市柯桥区西纺纺织产业创新研究院项目(19KQYB10)。

摘　　要：针对视觉词袋模型缺乏空间信息,对图像纹理表述不明确的问题,提出一种融合空间信息的词袋模型方法用于羊绒羊毛纤维识别。该方法通过融入局部特征间的相对位置特征,构造空间上下文近义词表,结合软分配的方式,将特征点分配给多个同义性较强的视觉单词,有效减弱了单词同义性和歧义性的影响。结合含空间信息的特征共同表征图像纹理,弥补了视觉词袋模型缺乏空间信息的缺点。实验结果表明,该方法相比于传统词袋模型的平均正确率提高了7.9%,其平均识别率可达93.3%,可用于羊绒羊毛纤维的自动分类识别。To solve the problem that the bag of visual words lacks spatial information in image representation,and unclear description of the image texture,a new method of bag of words model containing spatial information was proposed to identify cashmere wool fibers.The relative position information between local features was incorporated to construct a spatial context synonym table in this method.Combining with soft-assignment,feature points were assigned to several visual words with strong synonymy.The influence of synonymy and ambiguity between words is effectively reduced.Combining the features with spatial information to jointly represent the image texture,it makes up for the defect that the visual word bag model lacks spatial information.The experiment results demonstrate that the average accuracy of this method is improved by 7.9%compared with the traditional bag-of-word,and its average recognition rate can reach 93.3%,which can be used for automatic recognition and classification of cashmere and wool fibers.

关 键 词：羊绒羊毛 词袋模型 软分配 空间上下文近义词表 

分 类 号：TS102.3[轻工技术与工程—纺织工程]