Estimation Of Chlorophyll Content Distribution In Leaves Of Epipremnum Aureum Based On Optical Characteristic Parameters

Chlorophyll content,as one of the commonly used biochemical characteristics in plant phenotypes,is an important indicator for evaluating plant growth and health.Epipremnum aureum is an excellent ornamental plant,and it also has a great effect on improving indoor air pollution.Therefore,it is necessary to study the rapid non-destructive detection of chlorophyll content of Epipremnum aureum plants.At present,hyperspectral technology is often used for non-destructive detection of plant internal content,but the existing methods generally use simple modeling methods to associate hyperspectral information with chlorophyll,ignoring the absorption and scattering effects of light in plant tissues that can reflect its physiological characteristics.In this paper,combining hyperspectral and optical characteristic parameter data,an optical characteristic parameter inversion method based on hyperspectral images is proposed,which aims to improve the accuracy of non-destructive detection of chlorophyll content and at the same time provide new ideas for plant growth monitoring.The main research work of this article can be summarized as follows:Firstly,parameters measurement:hyperspectral images,optical characteristic parameters(absorption coefficient and scattering coefficient)and chlorophyll SPAD values of 250 leaf samples of Epipremnum aureum were obtained by using hyperspectral imaging system,double integrating sphere and SPAD measuring instrument,respectively.Secondly,the simulation modeling method is proposed to make up for the shortage of actual measurement data.Firstly,a single layer biological tissue model under parallel light was constructed based on Monte Carlo method to simulate the photon transmission inside the tissue.The model depth and other parameters were set according to the measured sample thickness and optical characteristic parameters,and different simulated photon distribution images were obtained respectively.The parallelization technology is used to realize the parallel computation of photon transmission,and a large number of noiseless simulation data can be obtained in a short time.Then use the convolutional neural network to classify and invert the optical characteristic parameters,use the hyperspectral images obtained by the simulation to train the neural network and adjust the network parameters to obtain the pre-training model.And compare and analyze the result of inversion model in different input ranges,normalization methods and different learning rate.The final inversion result of the convolutional neural network on the simulated hyperspectral data is:the inversion accuracy rate ofμ_a is90.15%,and the inversion accuracy rate ofμ_s is 87.55%.The inversion accuracy obtained by training the measured data in the same way is:absorption coefficient:78.33%,scattering coefficient:75.17%.From the results,it can be seen that the inversion accuracy of the measured hyperspectral image is significantly lower than that of the simulated data.Finally,based on the pre-trained model obtained from the simulation data,the model migration and feature migration are combined,and a CDAM-trans transfer learning method combining inter-class and inter-domain difference measurements is proposed to improve the transfer ability of the model.The final inversion result of the measured data is that accuracy rate ofμ_a is 89.35%and accuracy rate ofμ_s is 88.30%,compared with the method of using CNN and the traditional multi-label multi-classification algorithms to directly training measured data,the effect has been significantly improved.On this basis,the chlorophyll content of the Epipremnum aureum leaves was estimated.First,the hyperspectral,optical characteristic parameters,and depth features obtained by CDAM-trans method were used as the input of the regression model,and the chlorophyll SPAD value was used as the output to obtain the prediction model.It can be seen from the experimental results that the optical characteristics parameters have better prediction result than the hyperspectral data,indicating that the optical characteristics parameters can reflect the internal information of plant tissues.Compared with these two methods,the deep features obtained by transfer learning have the best prediction effect,which proves the correctness of the optical characteristics parameter inversion method.Finally,a model was constructed to predict the chlorophyll content of the entire leaf according to the deep features obtained from the CDAM-trans method,and the visualization of the chlorophyll content distribution was realized.