基于图像分析技术的烤烟上部叶采收成熟度判别  

Harvest maturity identification for upper flue-cured tobacco leaves based on image analysis technology

作　　者：路晓崇[1] 杨超[2] 王松峰[3] 鄢敏 杨洋[4] 彭玖华 郑小雨 杨懿德 LU Xiaochong;YANG Chao;WANG Songfeng;YAN Min;YANG Yang;PENG Jiuhua;ZHENG Xiaoyu;YANG Yide(College of Tobacco Sciences,Henan Agricultural University,Zhengzhou 450002,China;Chongqing Branch of China National Tobacco Corporation,Chongqing 400023,China;Key Laboratory of Tobacco Biology and Processing,Ministry of Agriculture,Tobacco Research Institute of CAAS,Qingdao 266101,Shandong,China;Yibin Branch of Sichuan Provincial Tobacco Company,Yibin 644000,Sichuan,China)

机构地区：[1]河南农业大学烟草学院,郑州市450002 [2]中国烟草总公司重庆市公司,重庆市400023 [3]中国农业科学院烟草研究所农业部烟草生物学与加工重点实验室,山东省青岛市266101 [4]四川省烟草公司宜宾市公司,四川省宜宾市644000

出　　处：《烟草科技》2021年第5期31-37,共7页Tobacco Science & Technology

基　　金：四川省烟草公司宜宾市公司资助项目“云烟116适应性及配套技术研究与示范”(201951150020107);中国烟草总公司重庆市公司资助项目“提高渝金香烟叶成熟度关键技术研究与应用”(NY20180601070002)。

摘　　要：为实现烤烟上部叶成熟度的客观准确判断,降低人为主观因素对烤烟采收成熟度判断的错误率,在烟叶采收前,对不同田间成熟度的烤烟上部叶图像进行采集,利用MATLAB2018b提取烤烟颜色特征值与纹理特征值,建立BP神经网络模型,并对烤烟采收成熟度进行分类甄别。结果表明,不同成熟度烤烟上部叶的颜色特征值与纹理特征值有较明显的差异,所建立的BP神经网络模型能够较为准确地识别不同成熟度的烤烟上部烟叶,其中训练样本的预测值与实际值的决定系数达到0.9855,验证样本的预测值与实际值的决定系数达到0.9819。所建立的基于烟颜色特征值与纹理特征值的判别烤烟上部叶成熟度的BP神经网络模型具有较好的判别效果。In order to discriminate the maturity of upper flue-cured tobacco leaves objectively and accurately and to reduce the error rate caused by subjective factors,the images of upper leaves with different field maturity were collected before harvesting,and the values of color characteristics and texture characteristics of flue-cured tobacco were extracted by MATLAB2018b.A BP neural network model was established to discriminate the harvest maturity of flue-cured tobacco.The results showed that the color characteristic values and texture characteristic values of upper flue-cured tobacco leaves of different maturity levels differed greatly.The established BP neural network model could accurately identify the upper leaves of different maturity levels.The coefficient of determinations of the predicted and actual values of the training samples both reached 0.9855,and those of the verification samples reached 0.9819.The established model based on color and texture characteristics was effective for discriminating the maturity of upper flue-cured tobacco leaves.

关 键 词：烤烟 上部叶 成熟度 BP神经网络 

分 类 号：S572[农业科学—烟草工业]