基于可见光机器视觉的棉花伪异性纤维识别方法  

Lint Cotton Pseudo-foreign Fiber Detection Based on Visible Spectrum Computer Vision

作　　者：王欣[1,2] 李道亮[1,3] 杨文柱[4] 李振波[1,5] 

机构地区：[1]中国农业大学信息与电气工程学院,北京100083 [2]中国农业大学网络中心,北京100083 [3]中国农业大学北京市农业物联网工程技术研究中心,北京100083 [4]河北大学数学与计算机学院,保定071002 [5]中国农业大学农业部农业信息获取技术重点实验室,北京100083

出　　处：《农业机械学报》2015年第8期7-14,共8页Transactions of the Chinese Society for Agricultural Machinery

基　　金：国家自然科学基金资助项目(31228016;61100115);农业科技成果转化基金资助项目(2012GB23600629);'十二五'国家科技支撑计划资助项目(2011BAD21B01;2012BAD35B07)

摘　　要：为提高皮棉质量和皮棉中异纤的检测精度,提出了一种基于机器视觉的棉花伪异性纤维识别方法。皮棉经过开松装置被制成薄棉层,检测通道两侧的相机对棉层进行拍摄,并将采集到的棉层及异纤和伪异纤图像保存到工控机,通过图像分块及阈值分割等算法,提取伪异纤目标区域,统计获取区域的数个颜色、形状和纹理特征,基于特征数据,分别使用BP神经网络、一对一有向无环图策略线性核函数支持向量机和径向基核函数支持向量机对两大类棉花杂质进行分类识别。实验结果表明,99.15%的伪异纤目标可被准确识别,径向基核函数支持向量机在棉花异纤和伪异纤分类识别中,总分类正确率为95.60%,能够满足在线检测的要求。The quality and level of lint cotton are degraded because there are many foreign fibers and other harmful non-fiber trashes which are mixed into it in the process of plantation, production, transportation and machining. It will bring direct economic loss to textile industry. In order to improve the quality of lint cotton and increase the detection rate of foreign fibers, a pseudo-foreign fiber detection method based on visible spectrum machine vision was proposed. Lint cotton was made of thin layer after opening, and then transferred to the detection passage. Images of cotton layer with foreign fibers and pseudo-foreign fibers were snapshot by two line-scan cameras installed by the side of detection passage, and then it was stored into the industrial personal computer＇s hard disk of experimental platform. Algorithms of image block and threshold were applied to extract pseudo-foreign fibers target areas, and statistical features in color, shape and texture of these target areas were calculated. Three classifiers ： BP neural network, one to one directed acyclic graph linear kernel SVM and RBF kernel SVM were used to separate the two categories of cotton impurities. Results showed that 99. 15 % of the pseudo-foreign fibers can be accurately identified, and the performance of RBF kernel SVM was the best among the three classifiers. With average recognition rate of 95.60%, the RBF kernel SVM can meet the online detection requirements of lint cotton trashes.

关 键 词：棉花异性纤维 识别 机器视觉 BP神经网络 支持向量机 

分 类 号：S24[农业科学—农业电气化与自动化] TP391[农业科学—农业工程]