基于数字图像处理的羊绒与羊毛识别研究  

Study on the Recognition of Cashmere and Wool Based on Digital Image Processing

作　　者：刘爽 许忠保[1] 李春桥 陈威[1] 宋丛珊 LIU Shuang;XU Zhongbao;LI Chunqiao;CHEN Wei;SONG Congshan(Hubei University of Technology,Hubei Wuhan,430068;Hubei Province Fiber Inspection Bureau,Hubei Wuhan,430060)

机构地区：[1]湖北工业大学,湖北武汉430068 [2]湖北省纤维检验局,湖北武汉430060

出　　处：《棉纺织技术》2019年第10期30-34,共5页Cotton Textile Technology

基　　金：湖北工业大学高层次人才启动基金项目(BSD2012002)

摘　　要：探讨基于数字图像处理的羊绒与羊毛纤维识别方法。利用螺旋相位相衬显微镜采集到特征清晰的羊绒与羊毛纤维图像,经图像预处理后,采用分段扫描法提取纤维直径,鳞片骨架法提取纤维高度等形态特征,并分别提取不同方向上的灰度共生矩的能量、熵、对比度、相关性等4个特征向量的均值和标准差作为纹理特征参数。最后应用BP神经网络,通过误差反馈调节,得到最佳参数的BP神经网络模型。试验表明:羊绒与羊毛纤维识别的正确率达到93.3%。认为:采用数字图像处理提取纤维特征能较好地识别羊绒与羊毛纤维。The recognition method of cashmere and wool fiber based on digital image processing was discussed.Spiral phase contrast microscope was used to collect the images of cashmere and wool fibers with clear characteristics.After the image pre-treatment,segmental scanning method was adopted to extract the fiber diameters.The morphological characteristics like the fiber height were extracted by the scale skeleton method.The mean value and the standard deviation of four eigenvectors including energy,entropy,contrast,correlation of the gray dependence matrix in different directions were extracted respectively and were taken as the textural feature parameters.In the end,BP neural network was adopted.By error feedback adjustment,the BP neural network model with the optimized parameters was obtained.The experiment showed that the accuracy rates of cashmere and wool fiber recognition were reached 93.3%.It is considered that the cashmere and wool fiber can be better recognized by adopting digital image processing to extract the fiber characteristics.

关 键 词：数字图像处理 羊绒 羊毛 纤维识别 螺旋相位相衬显微镜 灰度共生矩 BP神经网络 

分 类 号：TS101.8[轻工技术与工程—纺织工程]