开放环境下烟叶等级RGB图像智能识别及判别模型的构建  

Intelligent Recognition of Tobacco Grade by RGB Images and Establishment of Discriminant Model for Grading of Flue-cured Tobacco in an Open Environment

作　　者：童德文 陈钰[2] 杜超凡 雷佳[3] 范辉 林雷通 石三三 沈平 TONG Dewen;CHEN Yu;DU Chaofan;LEI Jia;FAN Hui;LIN Leitong;SHI Sansan;SHEN Ping(Wuping Branch Company,Longyan Tobacco Company,Wuping,Fujian 364300;Longyan Tobacco Company,Fujian Tobacco Company,Longyan,Fujian 364000;Guangdong China Tobacco Industrial Limited Company,Guangzhou,Guangdong 510310,China)

机构地区：[1]龙岩市烟草公司武平分公司,福建武平364300 [2]福建省烟草公司龙岩市公司,福建龙岩364000 [3]广东中烟工业有限责任公司,广东广州510310

出　　处：《贵州农业科学》2020年第3期131-135,共5页Guizhou Agricultural Sciences

基　　金：福建省烟草公司龙岩市公司科技与创新项目“烟叶初分等级快速诊断系统的研发”(2020Y)。

摘　　要：为烟叶生产及分级人员提供高效、精确、方便和低成本的烤烟等级智能判别方法,采用开放环境下烟叶RGB图像识别方法,通过百分位区间估计,确定B2F和C3F等级烟叶外观质量的红色(Red)、绿色(Green)、蓝色(Blue)3个色彩通道及灰度图像叶色均值区间及叶型特征区间,并基于贝叶斯分类器,构建智能判别模型,通过系统固化后对烟叶样品进行等级识别判定。结果表明:红色(Red)、绿色(Green)、蓝色(Blue)通道及灰度图像的色阶均值25%分位点及75%分位点可作为B2F和C3F等级烟叶叶色判别区间的上限和下限;烟叶叶片长宽比及叶片有效面积占比不满足判别区间的条件,不选择叶型参数作为判别区间;构建的模型对B2F和C3F等级烟叶判别的准确率分别为92.86%和95.56%,总体准确率为94.21%,且不受干扰叶片的影响,判别精度较高。Three color channels(red,green and blue),leaf color mean interval,leaf profile characteristics interval of grey-scale images of B2F and C3F tobacco are determined by estimating percentile interval according to tobacco RGB image recognition method in an open environment.The intelligent discriminant model is established based on Bayesian classifier and then the intelligent discriminant model after system solidification is used to recognize and discriminate the grade of tobacco samples to provide efficient,accurate,convenient and low-cost intelligent discrimination method of cured tobacco grading for tobacco production and grading staff.Result:Red,green and blue channels and color gradation mean value of grey-scale image with 25%quantile and 75%quantile can be used as the upper and lower limit of leaf color discriminant interval in B2F and C3F tobacco.The leaf profile parameter cannot be chosen as discriminant interval if leaf length-width ratio and effective leaf area ratio do not meet the condition of discriminant interval.The accuracy rate of B2F and C3F tobacco discriminated by the established model is up to 92.86%and 95.56%and the overall accuracy rate is 94.21%.The discrimination accuracy of the established model with high discrimination accuracy is not affected by the disturbance leaves.

关 键 词：RGB图像 烤烟等级 判别模型 智能判别 自动分级 开放环境 

分 类 号：S572[农业科学—烟草工业]