基于多纹理特征的白酒摘酒酒花图像分类识别  

Classification and recognition of liquor receiving hops images based on multiple texture feature extraction

作　　者：杨静娴 任小洪 YANG Jingxian;REN Xiaohong(Artificial Intelligence Key Laboratory of Sichuan Province,Sichuan University of Science and Engineering,Yibin 644000,China;School of Automation and Electronic Information,Sichuan University of Science and Engineering,Yibin 644000,China)

机构地区：[1]四川轻化工大学人工智能四川省重点实验室,四川宜宾644000 [2]四川轻化工大学自动化与信息工程学院,四川宜宾644000

出　　处：《包装与食品机械》2021年第4期38-45,共8页Packaging and Food Machinery

基　　金：四川省科技厅重点研发项目(2016SZ0074)。

摘　　要：根据白酒的“看花摘酒”传统手工摘酒经验,提出一种基于酒花视觉图像的多特征智能分类识别方法。通过对连续获取的酒花视频图像,在图像预处理基础上,提出采用局部二值模式(LBP)与灰度共生矩阵(GLCM)分别进行纹理特征提取,并提出LBP+GLCM相结合的纹理特征提取算法。通过对比旋转不变模式LBP、等价模式LBP、旋转不变等价模式LBP,确定对白酒酒花特征描述提取效率最高的等价模式LBP;对GLCM提取的特征值计算均值,并采用不同的特征值组合方式作为支持向量机(SVM)的输入得到分类结果;最后对得到的LBP特征与GLCM特征值进行特征级融合作为分类器的输入,并利用3种不同核函数的SVM分类器进行训练和测试。试验结果表明,LBP+GLCM分类准确率相较于单一LBP、GLCM均有不同程度地提高,其稳定性也高于单一特征分类。According to the traditional manual liquor receiving experience of liquor-receiving according to liquor hop,a multifeature intelligent classification and recognition method based on visual image of hops was proposed.Through the continuous acquisition of hop video images,based on image preprocessing,it was proposed to use local binary pattern(LBP)and gray level co-occurrence matrix(GLCM)to extract texture features respectively,and the extraction algorithm of texture features in combination with LBP+GLCM was proposed.By comparing the rotation-invariant mode LBP,the uniform mode LBP,and the rotation-invariant uniform mode LBP,the uniform mode LBP with the highest extraction efficiency for the liquor hop feature description was determined;By calculating the mean value of the feature values extracted by GLCM,and by using different combination of feature values as the input of Support Vector Machine(SVM),the classification result was obtained;finally,feature-level fusion of the obtained LBP features and GLCM feature values were used as the input of the classifier,and SVM classifier with three different kernel functions was used for training and test.The experimental results show that the accuracy of LBP+GLCM classification was higher than that of single LBP,GLCM,and its stability was higher than that of single feature classification.

关 键 词：多纹理特征 局部二值模式 灰度共生矩阵 酒花图像分类 

分 类 号：TS206.4[轻工技术与工程—食品科学]