Study On The Hyperspectral Characteristics Of Rami E Leaves And The Establishment And Application Of Qualitative And Quantitative Models

Nondestructive optical detection and diagnostic techniques has become a research hotspot in agricultural field,based on spectral remote sensing technology,using computer data analysis and processing as auxiliary means.Ramie?Boehmeiria nivea L?is a special and traditional fiber crop in China,has higher economic status.Comprehensive understanding of the hyperspectral response characteristics of ramie,establishing the hyperspectral ramie pattern recognition model,exploring the quantitative relationship between hyperspectral and physiological parameters of ramie leaves,will helpful for the ramie cultivation,germplasm resources development and utilization,and provide key technical supports for implementing the high yield and quality of ramie and the precision management of ramie farmlands,and significant for improving ramie yield and quality;and has wider application prospect in promoting the sustainable development of the ramie cause.This article is based on the Ramie Germplasm Resource nursery of Changsha County plum blossom base,the ramie garden of the Hunan Agricultural University and the ramie in the long-term location test point of the national hemp of Hunan Agricultural University.Under the condition of field cultivation,the hyperspectral characteristics of 40 varieties of ramie leaves were studied on the basis of spectral analysis technology.The hyperspectral identification model of 9 ramie varieties,the hyperspectral recognition model of ramie brown spot disease and the hyperspectral prediction model of the leaf water content of ramie were established.The primary research works and conclusions are showed as follows:1.The hyperspectral data of 40 different genotype ramie leaves,include brown spot leaves and healthy leaves,leaves at upper,middle and lower positions,were statistically analyzed from the respects of the original hyperspectral peak and trough parameters,the sub-band skewness and kurtosis parameters,and the three-edge parameters.The following hyperspectral characteristics of ramie leaves were fond:?1?The hyperspectrum differences of different ramie genotypes were mainly reflected in the reflectivity and area within the visible spectrum?400⁷80nm?or the shortwave infrared spectrum?1350²400nm?.The reflectivity and position showed obviously differences in the near-infrared band?780¹350 nm?.Among the parameters,the differences in the position of the 5^th trough,the reflectivity at the 2^nd peak,the position of the yellow edge,the amplitude at the red edge,the kurtosis at n6 and the skewness at n4 were smaller;but the parameters,such as the position of 1^st peak,the reflectivity at the green peak,the position of the red edge,the amplitude at the blue edge and the kurtosis at n4,the differences were bigger.?2?In the visible spectrum,the reflectivity of hyperspectrum at the lower position leaves was the highest and was inversely proportional to the SPAD value.While in the spectrum of 780¹350 nm,the reflectivity of different position leaves were in the same order,etc.the upper position>the middle position>the lower position,and were negatively correlative with the water contents.The reflectivity differences were not obvious in the spectrum of1350²450 nm,the upper position leaves had the lowest water content and also had the lowest reflectivity.For three-edge parameters,the positions of the blue and the yellow edges of the upper,middle and lower position leaves were more stable than that of the red edges,and their amplitude and area were also much smaller.The red edges of the middle position leaves,which had the largest SPAD values,were more close to the direction of long wave,while the lower position leaves with the smallest SPAD value were more close to the direction of short wave.?3?To compare the hyperspectral curves of the brown spot disease leaves and healthy leaves,there were obvious differences in the reflectivity of the green peak region,the position of the 2^nd trough,the skewness at n3,the kurtosis at n4 and the amplitude of the blue edge;and in the range of 670-970nm,the reflectivity of the healthy leaves were obviously higher than that of the brown spot disease leaves.2.Nine recognition models for recognizing ramie varieties and leaf's brown spot disease were established,base on the original hyperspectral peak and trough parameters,the skewness and kurtosis parameters,the three-edge parameters and principal component analysis,discriminant analysis?DA?and the support vector classification?SVC?.The research results showed that,the model including twenty principal components and the SVC?the RBF kernel function?was the best in the nine ramie variety recognition models,it has the fewest variables,and its correct rate reached 96.91%.While in the recognition models of ramie leaf's brown spot disease,the best effect was achieved by the model including nine principal components and the mahalanobis distance discriminant analysis as well as with nine principal components and the SVC?the RBF kernel function?,the correct rate of its prediction set reached 100%.3.A variety of pretreatment methods were used to reduce the original hyperspectral noise,then,the correlation coefficient method,the regression coefficient method,the correlation-regression coefficient method and the regression-regression coefficient method were used respectively to extract the characteristic bands and the characteristic wavelengths.And in the end,the partial least squares regression?PLSR?and the support vector regression?SVR?were adopted respectively to establish the prediction model about ramie leaf's water content based on full band,the characteristic band and the characteristic wavelength.The results were showed:the effect of the Savitzky-Golay smoothing-PLSR model was the best,when the full band was used as the input variable,its variable number was 2031,R²=0.7164 and RMSEP=0.0292.The effect of the normalize-regression coefficient band-PLSR model was the best,when the characteristic band was used as the input variable,its variable number was 360,R²=0.7153 and RMSEP=0.0292.The effect of the SNV-regression-regression coefficient band-PLSR model was the best,when the characteristic wavelength was used as the input variable,its variable number was 12,R²=0.7016 and RMSEP=0.0299.The hyperspectral characteristics of ramie leaves was studied intensively from various perspectives,including the original hyperspectral peak and trough parameters,the sub-band skewness and kurtosis parameters and the three-edge parameters etc..Based on the research results,the prediction models for ramie variety recognition and brown spot diseases of ramie leaf,and prediction model for water content of ramie leaf were established,these fill the current research blank.Moreover,the original hyperspectral peak and trough parameters,the sub-band skewness and kurtosis parameters were selected as the feature variables to build the qualitative recognition models,and the impacts of the principal components on the model effects were discussed deeply.In establishing the quantitative prediction models,the methods based on correlation coefficient,the secondary extraction of characteristic wavelength of the regression coefficient characteristic waveband were proposed,these are the innovations of this study.