Research On Hyperspectral Remote Sensing Technology For Crop Monitoring

In this thesis,the hyperspectral remote sensing technology in crop monitoring is taken as the research object,mainly studies the mechanism of crop hyperspectral remote sensing monitoring,biophysical and chemical characteristics of crops and hyperspectral data processing and modeling methods;based on the correlation of hyperspectral image bands,efficient and low computational complexity feature extraction methods have been studied;hyperspectral image classification algorithm and experimental research;hyperspectral feature extraction algorithm and classifier matching strategy;hyperspectral data and crop biological parameters of the response relationship,extraction of sensitive to crop biological parameters of the characteristic spectrum,spectral index and biological parameters correlation analysis,construction an inversion model for crop biological parameters.The main conclusions and innovations of this thesis are as follows:(1)In this thesis,a segmented minimum noise fraction(SMNF)transformation is proposed for efficient feature extraction of hyperspectral images(HSIs).The original bands can be partitioned into several highly correlated subgroups based on the correlation matrix image of the hyperspectral data.The MNF is implemented separately on each subgroup of the data,and then,the Bhattacharyya distance is used as the band separability measure for feature extraction.Consequently,the extracted features can then be significantly classified using state-of-art classifiers,i.e.,k-NN or SVM.Experiments on two benchmark HSIs collected by AVIRIS and ROSIS demonstrate that the proposed method significantly reduces the transformation time in comparison with the conventional MNF.The results indicate that the computation time of proposed SMNF is reduced by 62.5% for Indian Pines,and 41.3% for Pavia University scene,respectively.The Fisher scalars' criterion shows that the class separability with the segmented MNF is the best,and the extracted features based on SMNF even exhibit higher classification accuracy in separating crop species compared with the MNF.(2)To reduce the training time of SVM for a large dataset,a support vectors(SV)pre-extraction method based on a radial basis function neural network(RBFNN)is proposed in this thesis.Because the RBFNN has good convergence and fast training,the proposed method seeks to obtain an optimal decision boundary based on the RBFNN and then to extract a candidate set of SVs from a large dataset using the decision boundary.In the present design of the RBFNN,the adaptive k-means algorithm for clustering computation is applied,followed by recursive least squares(RLS)for computing the weight vector.Thereafter,a candidate set can be extracted from the original training set using the decision boundary.It is worth noting that the candidate set is far smaller than the original training set.The experimental results show that the proposed method improves the performance of the SVM,which ensures its generalization ability and classication speed.(3)An estimation model based on hyperspectral remote sensing technique is proposed to estimate the chlorophyll content of Gannan navel orange in this thesis.The spectral data of 100 navel orange leaf samples from 10 navel orange trees in Gannan navel orange planting demonstration farm were obtained by spectral analyzer with 1.67 nm and 3.41 nm output band interval and spectral range of 400-1000 nm.The chlorophyll content of each navel orange leaf was detected by spectrophotometry.The estimation model was established based on partial least squares regression(PLSR)algorithm using original spectrum,first derivative spectrum and second derivative spectrum.The results showed that the chlorophyll content of navel orange leaves was sensitive to 400-780 nm,and the prediction model based on PLSR was effective.The prediction effect of chlorophyll content of navel orange leaves with 1.67 nm band interval was better than that with 3.41 nm band interval.When the original spectrum is used to model,the prediction of chlorophyll a,chlorophyll b and total chlorophyll content is the best by using the full band and five principal components,the correlation coefficients for chlorophyll a,chlorophyll b and total chlorophyll content were 0.97,0.95 and 0.98,C-RMSE 0.03,0.05 and 0.05,V-RMSE 0.15,0.06 and 0.19,respectively.Compared with the original spectral model,the first derivative spectral model can achieve better prediction results only at low principal component number(1-4),but at high principal component number(5-10),the prediction results are worse than the original spectral model.The predicted correlation coefficients of chlorophyll content were more balanced in all principal component numbers by second derivative spectroscopy modeling,it achieved high prediction accuracy,especially in the prediction of chlorophyll b,compared with the original spectrum and first derivative spectrum,the advantage is obvious.In general,the second derivative,1.67 nm band interval and full band have the best modeling effect.(4)In this thesis,a method based on hyperspectral remote sensing technology to detect the acidity and vitamin C content of Gannan navel orange fruit was proposed.Spectral data of 30 navel orange samples from a navel orange plantation in Xinfeng County,Jiangxi Province,were obtained by spectral analyzer with a spectral range of 900-1700 nm.The contents of titratable acid and vitamin C in samples were determined by chemical analysis of sodium hydroxide solution and potassium iodate solution,respectively.The results showed that the internal quality of navel orange fruit was sensitive to 900-1700 nm.The partial least squares regression was used to model the titratable acid at 950-1350 nm,the optimum principal component was 4,the correlation coefficients of training set and test set were 0.9767 and 0.9085,respectively.The content of vitamin C was modeled at 900-1600 nm,the optimum principal component was 6,R2 of training set and test set were 0.9387 and 0.8690,respectively.The model has good prediction effect and can provide technical support for batch testing of the internal quality of Gannan navel orange.In this thesis,some new hyperspectral remote sensing methods and models are proposed for crop species identification,chlorophyll content estimation and internal quality detection of agricultural products,and high precision is achieved.The research results can provide theoretical basis and technical support for precision agriculture,yield estimation and rapid quality detection of agricultural products,and promote highlight.Its can promote the engineering application of hyperspectral remote sensing technology in crop planting and production.