基于叶色偏态分布模式的鲜烟叶成熟度判定

Maturity discrimination on fresh tobacco leaves based on skewed leaf color distribution patterns

为基于烟叶外观表型对鲜烟叶成熟度进行智能化判定,在拓展叶片RGB(Red Green Blue)颜色模型图像色彩信息参数的基础上,构建并比较分析了不同参数体系、建模方式下的鲜烟叶成熟度的判定模型。结果表明:①不同成熟度的鲜烟叶叶色偏态参数间存在显著差异,且不同叶色信息参数随着成熟度的提高呈现不同的变化规律;②采用反向传递神经网络(Back Propagation Neural Network,BPNN)构建的鲜烟叶成熟度判定模型的拟合度均优于采用多元逐步回归法。同时,叶色偏态参数可明显提高BPNN模型对鲜烟叶成熟度的判定准确度,其中基于叶色偏态参数构建的BPNN模型综合表现最优,尤其是在不同生产年份的鲜烟叶成熟度预测方面,模型表现出明显优势。因此,采用基于叶色偏态参数的BPNN模型可对鲜烟成熟度进行准确判定。

In order to discriminate the maturity of fresh tobacco leaves intelligently,based on the extension of color information parameters of RGB(Red Green Blue)color model,discrimination models for fresh tobacco leaf maturity were developed by adopting different parameter systems and modeling methods.The results showed that:1)There were significant differences among the leaf color skewness parameters of fresh leaves with different maturity levels,and different leaf color information parameters changed differently with the increase of maturity.2)The fitting degrees of the discrimination models developed by BPNN(Back Propagation Neural Network)method were superior to those by a multiple stepwise regression method.Meanwhile,the adoption of the skewness parameters significantly improved the discrimination accuracy of BPNN models for the maturity of fresh leaves,among which the BPNN model based on the skewness parameters of leaf color had better integrative performance.Therefore,the BPNN model based on the skewness parameters of leaf color could be used to discriminate the maturity of fresh tobacco leaves.