机器视觉的烟叶物理特性表征方法

为了研究梅州地区各等级烟叶的外观特征与物理特性间的关系,找到一种通过烟叶外观特征表征其物理特性的方法,选取了梅州6个产地、12个等级的初烤烟叶共977片。使用机器视觉设备和质构仪分别检测了烟叶样本的外观特征和物理特性。选取其中781片烟叶样本作为训练集,使用了弹性网络、极端随机树、支持向量机等回归模型以及模型融合技术分别构建了基于烟叶外观特征的最大拉力、剪切力和撕裂度的表征模型。选取196片烟叶样本作为测试集,以平均绝对误差为模型评价指标,评估了3种表征模型的泛化性能。结果表明,对于最大拉力的表征模型而言,模型在测试集上的预测值与真实值的相关系数超过0.73,拟合优度为0.54;对于剪切力的表征模型而言,模型在测试集上的预测值与真实值的相关系数超过0.78,拟合优度为0.60;对于撕裂度的表征模型而言,模型在测试集上的预测值与真实值的相关系数超过0.75,拟合优度为0.56。烟叶的外观特征对于烟叶的最大拉力、剪切力和撕裂度具有一定的表征能力。

Characterization Method of Tobacco Leaves Physical Properties Based on Machine Vision

In order to study the relationship between the appearance characteristics and the physical properties of various grades of tobacco in Meizhou, an attempt was made to find a method about characterizing the physical properties of tobacco through its appearance characteristics. A total of 977 pieces of first-roasted tobacco leaves of 12 grades from six origins in Meizhou were selected. The appearance characteristics and physical properties of the tobacco samples were examined using machine vision equipment and texture analyzer, respectively. A total of 781 tobacco samples were selected as the training set. Regression models such as Elastic Net, Extremely Randomized Trees and Support Vector Machine were used along with the Ensemble technique to construct the characterization models of maximum tensile force, shear force and tearing degree based on the appearance characteristics of tobacco samples. A total of 196 tobacco samples were selected as the test set, and the generalization performance of the three characterization models were evaluated using the mean absolute error. The results indicated that the maximum tensile force model exhibited a correlation coefficient over 0.73 between the predicted and true values of the samples in the test set, with a goodness of fit of 0.54. Similarly, the shear force model demonstrated a correlation coefficient exceeding 0.78 and a goodness of fit of 0.60. Additionally, the tearing degree model displayed a correlation coefficient surpassing 0.75 and a goodness of fit of 0.56 for the predicted and true values of the samples in the test set. The appearance characteristics of tobacco leaves have certain ability to characterize the maximum tensile force, shear force and tearing degree of tobacco leaves.