| Hyperspectral imaging technology can simultaneously collect spatial information and spectral information of crops,which is an important method for obtaining information about crop growth.It has important prospects for application in monitoring the growth status of horticultural vegetables.However,with the rapid development of hyperspectral imaging technology,the processing of hyperspectral image data is increasingly becoming a difficult problem.Applying modern computer technology such as machine learning to hyperspectral data processing has become a hot topic in this field.However,there are two critical problems in this type of research.First,there lack proper method for analyzing and preprocessing data features before data modeling and the accuracy of the results entirely depends on the regression models or classification models,which may result in serious errors in the prediction results.Second,different research objects have different spatial and spectral features,and there is no universal modeling method,which increases the difficulty of modeling and classifying algorithms.To circumvent the problems above,this thesis takes the purple butter lettuce cultivated under five different light conditions in the facility environment as research project.Combining hyperspectral technology and machine learning technology,a regression model between the hyperspectral reflectance data and the measured relative chlorophyll content of lettuce leaves at three growth stages,i.e.tender leaf stage,growth stage and mature stage,is constructed,and classification models of the hyperspectral reflectance data of lettuce leaves are built.Specific research work is as follows:Firstly,this paper establishes the regression models of lettuce hyperspectral data and the measured relative chlorophyll content value based on the local weighted regression algorithm and neural network.With the hyperspectral reflectance data of purple butter lettuce at different stages which are illuminated at different conditions as input,the first derivative and the correlation with the measured chlorophyll relative content are calculated and the reflectance data are regrouped.Then,the local weighting regression algorithm and back propagation neural network algorithm are respectively used to establish the regression model of the hyperspectral data and the relative chlorophyll content.The results of modeling experiments show that the linear model of the test set for back propagation neural network has lower error and higher determination coefficient for the hyperspectral data of lettuce in each stage than that of local weighted regression algorithm.The linear model error reaches the lowest value of 1.67 at the growth period for the test set and the determination coefficient reaches the highest value of 0.98 at the young period.This thesis shows that the neural network modeling method is more suitable for hyperspectral data modeling,and has the characteristics of small error and high determination coefficient.Secondly,a hyperspectral data classification model of lettuce leaves based on K-nearest neighbor algorithm and support vector machine is constructed.The results of classification experiment show that the support vector machine has higher classification accuracy for hyperspectral data under different illumination conditions at each stage,with accuracy of training set reaching 90.5% at maturity period and an accuracy of 86% at the tender leaf period.It indicates that support vector machine can be used to classify the hyperspectral experimental data of lettuce leaves more effectively and reveals that different light combinations have a greater impact on the growing process of lettuce.These results are basically consistent with the chemical detection results of lettuce ripening.In summary,this thesis combines hyperspectral technology and machine learning technology to carry out modeling and classifying research on the hyperspectral data of lettuce.The results in this thesis provide a reference for light combination in greenhouse vegetable culture and an approach to monitor the growth status information of facility vegetables.It also reveals scientific basis for hyperspectral machine vision and its application in agriculture. |
