皮棉表面多类异性纤维的高光谱图像检测  

Detection of foreign materials on the surface of ginned cotton by hyper-spectral imaging

作　　者：郭俊先[1,2] 应义斌[1] 成芳[1] 康玉国 李付堂 饶秀勤[1] 

机构地区：[1]浙江大学生物系统工程与食品科学学院,杭州310029 [2]新疆农业大学机械交通学院,乌鲁木齐830052 [3]北京中棉机械成套设备有限公司,北京100089

出　　处：《农业工程学报》2010年第10期355-360,共6页Transactions of the Chinese Society of Agricultural Engineering

基　　金：国家科技支撑计划子课题(2006BAD11A14-1-1);国家自然科学基金项目(30825027);国家自然科学基金项目(61005022)

摘　　要：为了考察高光谱成像技术检测多类异性纤维共存时的性能,也为其今后用于商业化皮棉质量评价,或在线皮棉杂质分拣提供研究基础,试验采集丙纶丝和毛发共存的样本高光谱图像。依据单一异性纤维分割的最佳波段图像,获得灰度平均和小波变换后融合图像。通过对比小波变换融合图像,灰度平均图像,以及全波段下主成分和独立成分得分图像,获得用于目标分割的最佳图像。采用最优特征集和分类树判别方法,判别最佳图像分割后的连通区域,剔除噪声点和假阳性。依据目标判定原则,训练集和独立验证集的异性纤维识别率分别为84.09%和75.86%。结果表明,高光谱图像能够检测多类共存的异性纤维,验证集中灰色、白色丙纶丝和黑色毛发识别率为100%,白色毛发识别率为0。In order to investigate the potential for detecting multi-foreign materials on the surface of ginned cotton by hyper-spectral imaging,and also to provide the research basis for applying this imaging to evaluate cotton quality or to sort the trashes in line,cotton hyper-spectral images with multi-foreign materials were acquired in the reflection mode.Optimal waveband images for the detection of each foreign material were transformed into fusion images using wavelets transformation or averaging operator,respectively.The optimal images for the ultimate segmentation of foreign materials were selected from four types of the best potential principal components scores image,independent components scores image,fusion images of gray averaging and wavelets transformation.The classification tree and optimal subset of region features were developed to classify each region of binary images into the groups of foreign-objects or non-foreign-objects,so as to eliminate all the noise and the false positives.The recognition rates of foreign materials by using the discriminate rule of targets in the training and testing sets were up to 84.09% and 75.86%,respectively.The results show that the hyper-spectral imaging system in line scanning mode has the potential for the detection of coexistence foreign materials.In the testing set,all of black hairs,gray and white polypropylene fibers were detected while none of white hairs was detected.

关 键 词：图像识别 小波变换 分类 高光谱图像 皮棉 异性纤维 检测 

分 类 号：TS101.3[轻工技术与工程—纺织工程] TP181[轻工技术与工程—纺织科学与工程]