基于光学特性参数反演的绿萝叶绿素含量估测研究

Estimation of Chlorophyll Content of Epipremnum aureum Based on Optical Characteristic Parameter Inversion

为快速准确检测植物体叶绿素含量,提出一种基于MMD迁移的光学特性参数反演方法。以绿萝叶片为研究对象,仿真光子在基于蒙特卡洛方法的单层平板模型上的运动轨迹,获得12 000幅绿萝叶片仿真光亮度分布图,利用卷积神经网络对模拟光谱数据进行训练,得到预训练模型;基于预训练模型进行迁移学习,在少量实测绿萝叶片光谱数据上对模型进行微调,进行绿萝光学参数反演,得到吸收系数μa反演准确率为84.83%、散射系数μs反演准确率为83.33%;在此基础上引入最大均值差异方法,提升迁移效果。结果表明,与普通的模型迁移方法相比,基于MMD迁移的方法具有更好的反演效果,吸收系数μa反演准确率为87.55%,散射系数μs反演准确率为86.67%。利用MMD迁移得到的全连接层特征建立叶绿素回归模型的决定系数R~2为0.931 0,分别比直接使用光学参数和光谱图像建立的模型决定系数R2高0.046 8和0.062 0。研究表明,基于光学特性参数反演方法可以为叶绿素含量无损估测研究提供参考。

In order to realize the rapid and accurate detection of chlorophyll content in plants,an inversion method based on MMD migration was proposed.Taking Epipremnum aureum leaves as the research object,the motion trajectory of photons was simulated with the Monte Carlo method based single-layer flat plate model,totally 12000 simulated luminance distribution maps were obtained.The convolutional neural network was used to train the simulated spectral data to obtain the pre-training model.Then based on the pre-training model,the model was fine-tuning on the measured spectral data of a small amount of Epipremnum aureum leaves to realize the inversion of the optical parameters.The inversion results were as follows:absorption coefficientμa was 84.83%and scattering coefficientμs was 83.33%.On this basis,the maximum mean difference method was added to improve the migration effect.The results showed that the MMD migration method had a better inversion effect with absorption coefficientμa was 87.55%and scattering coefficientμs was 86.67%compared with the common model migration method.The chlorophyll regression model was established by using the full connection layer characteristics obtained from MMD migration,and the determination coefficient R 2 of this method was 0.0468 and 0.0620 higher than that of the model established directly using optical parameters and spectral images,respectively.The experimental results showed that the inversion method based on optical characteristic parameters can provide important reference for the research of chlorophyll nondestructive detection.