基于高光谱成像的烤烟着生部位识别  

Growth position discrimination of flue-cured tobacco leaves based on hyperspectral imaging

作　　者：梅吉帆 郭文孟 李智慧 薛宇毅 杨忠泮 李嘉康 苏子淇 张雷 堵劲松[1] 徐大勇[1] 李辉 MEI Jifan;GUO Wenmeng;LI Zhihui;XUE Yuyi;YANG Zhongpan;LI Jiakang;SU Ziqi;ZHANG Lei;DU Jingsong;XU Dayong;LI Hui(Key Laboratory of Tobacco Processing,Zhengzhou Tobacco Research Institute of CNTC,Zhengzhou 450000,China;Zhanjiang Cigarette Factory,China Tobacco Guangdong Industrial,Zhanjiang 524033,China;Technology R&D Center,Gansu Tobacco Industrial Co.,Ltd,Lanzhou 100034,China;College of Electrical and Information Engineering,Zhengzhou University of Light Industry,Zhengzhou 450001,China)

机构地区：[1]郑州烟草研究院烟草工艺重点实验室,郑州450001 [2]广东中烟工业有限责任公司湛江卷烟厂,湛江524033 [3]甘肃烟草工业有限责任公司,兰州100034 [4]郑州轻工业大学电气信息工程学院,郑州450001

出　　处：《中国烟草学报》2024年第3期51-60,共10页Acta Tabacaria Sinica

基　　金：福建中烟科技项目“卷烟产品及原料高光谱特征分析与应用技术研究”(No.D2020248)。

摘　　要：【目的】采用高光谱成像技术结合机器学习方法,建立烤烟着生部位(上部、中部、下部)的识别模型。【方法】首先,通过分析烟叶在水、氮敏感波段下的强度分布特征,采用了一种结合OTSU和Sauvola图像分割算法的双阈值感兴趣区(ROI)选取方法,然后对比分析不同预处理方法对数据建模的影响规律,采用支持向量机(SVM)、极限梯度提升(XGBoost)算法进行判别模型的建立,通过参数寻优进行模型的优化。使用遗传算法(GA)和遗传算法结合连续投影算法(GA-SPA)进行特征波长的选择,建立简化模型。【结果】(1)建立的双阈值感兴趣区选取方法能准确高效地实现烤烟叶片正常叶面区域的选取(2)不同数据预处理方法对识别模型影响较为显著,基于一阶导和萨维莱茨-戈莱平滑(1Der+SG)预处理光谱数据,结合GA选取的特征波长建立的XGBoost着生部位识别模型具有最佳的分类效能,其准确率高达97.78%。【结论】研究建立的基于高光谱成像技术结合机器学习方法的部位模型可满足烤烟着生部位的高效准确识别。[Background]Using hyperspectral imaging technology combined with machine learning methods,a recognition model for the growth potion(upper,middle,lower)of flue-cured tobacco was established.[Methods]The intensity distribution characteristics of tobacco leaves in water and protein sensitive bands was analyzed firstly.Then,a two-threshold region of Interest(ROI)selection method combining OTSU and Sauvola image segmentation algorithms was raised based on the analysis of abundance distribution.Moreover,the influences of different data preprocessing methods on data modeling were comparatively analyzed.Support vector machine(SVM)and extreme gradient boosting(XGBoost)algorithm were adopted to establish the discrimination model,and the models were optimized by parameter optimization subsequently.Using the Genetic Algorithm(GA)and the combination of Genetic Algorithm with Successive Projection Algorithm(GA-SPA)for the selection of characteristic wavelengths,a simplified model was established.[Results]Test data show:1)The established dual threshold method can efficiently select the right leaf area of flue-cured tobacco leaves;2)the efficacy of discrimination model was obviously affected by data preprocessing methods.Based on the spectral data preprocessed using the first-order derivative and Savitzky-Golay smoothing(1Der+SG),combined with the characteristic wavelengths selected by GA,the XGBoost recognition model for the growth position had the best classification performance,with an accuracy rate of up to 97.78%.[Conclusion]The model established in this study,based on hyperspectral imaging technology combined with machine learning methods,can efficiently and accurately identify the growth position of flue-cured tobacco.

关 键 词：高光谱成像技术 着生部位 数据预处理 机器学习 双阈值分割 定性判别 

分 类 号：S572[农业科学—烟草工业] TP391.41[农业科学—作物学] TP18[自动化与计算机技术—计算机应用技术]