| Hyperspectral techniques promotes best remote sensing to go ahead by its big datum quantity, high distinguishing rate and strong continuity since its naissance in 1980s. Now, it has been applied abroad in resource survey, geological survey, atmosphere survey, disaster environment inspection, soil survey, city environment survey, geological survey, and has acquired better appliance results and economical benefit, so that it was applied gradually in crop growth inspection, crop disease and insect pest inspection, crop quality inspection, and so on. Hyperspectral characteristic capability, specially the bigger advantages in identifying and classifying the matters of earth's surface, picking-up avail information than other techniques, made this technique have huge application potential in vegetation precision classifying, crop growth inspection, field water and fertilizer condition inspection.Tobacco is an important economy crop in China, and its planting area is about one million hectares each year. Presently, basic tobacco field protection system is set up in many provinces, and big area planting, unite administer has already been a kind of current. However, management only by sense and experience is a bottle-neck to tobacco cultivation in our country now, so many agriculture operator couldn't obtain the tobacco growth information exactly, quickly, and without defacing, so that they couldn't manage tobacco purposefully. Hyperspectral remote sensing is a very good technique to resolve these problems. Correlative researches focused on rice, corn, wheat, cotton, and so on, but little was on tobacco. Flue-cured tobaccos are broad leaves and the harvest aims are also leaves, which have essential differences with above crops that are narrow or little leaves and flowers or fruits are harvest outputs. Therefore, the research achievements about using hyperspectral technique to study tobacco biophysical and biochemical indexes, identifying yields and qualities, pick up information quickly, have important meanings to improve inspection and manage levels of tobacco production and apply precision agriculture in tobacco cultivation.We set up treatments of different tobacco types, flue-cured tobacco varieties, N P K levels in this experiment, and used ASD Field spec FR 2500 spectral instrument to get hyperspectral data. This paper researched the relationships between 27 hyperspectral indexes of tobacco leaf, canopy and leaf biophysical, biochemical indexes in different growth conditions and periods, and the relationships between 27 hyperspectral indexes of tobacco canopy, fresh leaves, flue-cured leaves and chemical qualities, mineral elements, aroma compositions in flue-cured leaves by correlation analysis. Then, we established the estimating models of many biophysical, biochemical indexes and mineral elements in upper leaves, cutters, lower leaves, the whole plant leaves by stepwise regression. In the same method, we established the predicting models of chemical qualities, mineral elements, aroma compositions in flue-cured leaves. The models had good predicting effects through significant test to R~2 and independent regression coefficients, and test of the predicting values and measured values for checking samples. So the results were as follows: First, the hyperspectral curves of tobacco leaves and canopies appeared the same trends with the typical vegetation spectra, and the red edge position had blue transfer and red transfer with the change of growth periods. This experiment connected the red edge parameters of tobacco leaves with leaf maturity. The result showed that the red edge position could estimate leaf maturity very well. The leaf maturity was all good when its red edge position lied in 693-695nm in different N levels. It could provided harvest information quickly and exactly through inspecting the red edge position.Second, through analyzing the spectral reflectance of tobacco leaves in the room, the characteristic variable of total nitrogen, total chlorophyll a, total chlorophyll b, total carotenoid, chlorophyll a, chlorophyll b, carotenoid, was all SDr/SDb. The characteristic variable of fresh weight was SDr. The characteristic variable of dry weight was (SDr-SDb)/(SDr+SDb). The characteristic variable of fresh weight wasλg.Third, we analyzed nicotine solely for it was a special content of tobacco. The stepwise regression models of hyperspectral reflectance, first derivative spectra, spectral characteristic variables were established. Through analyzing the R~2 and independent regression coefficients of 3 regression equations and the correlation between estimating values and measured values, the results were that the model of first derivative spectra had best estimating effect, and the model of spectral characteristic variables had worst estimating effect, and the model of spectral reflectance had the middle effect. The correlation coefficient of estimating values and measured values was 0.998 from the model of first derivative spectra, and it was 0.806 from the model of spectral reflectance, and it was 0.256 from the model of spectral reflectance. The first two reached 0.01 significant level, and the third didn't reach 0.05 significant level.Fourth, Rg/Rr was screened out as the characteristic variable of LAI, AFW, ADW through many statistical methods. Not only Rg/Rr has the linearity correlation with them, and the correlation coefficients all reached 0.01 significant level, but also the regression equation whose first independent was also Rg/Rr had good estimating effects. So this result provided new index to analyze tobacco biological characters by remote sensing methods.Fifth, the close relationships between canopy spectral parameters and physiological indexes, mineral elements and the models which were established and tested showed that it was feasible to estimate physiological indexes, mineral elements by spectral parameters. There were many differences between the coefficients and models of different positions, because physiological indexes, mineral elements in different positions had diversities, but the spectral parameters were changeless. However, the correlations between estimating values and measured values of each position leaf physiological indexes, mineral elements reached 0.05 significant level, so it suggested that estimating values were precise and people could selected estimating model by needs.Sixth, the correlations between canopy spectral parameters, fresh leaf spectral parameters, flue-cured leaf spectral parameters and chemical quality, mineral elements, neutral aroma compositions and the models which were established and tested by checking samples showed that there were certain relationships between them, and the chemical quality, mineral elements, neutral aroma compositions could be exactly predicted by different spectral parameters. Although tobacco quality was quite complicated, and many factors including growth condition and curing course etc had great influence on them, but it was possible to predict them simply by hyperspectral parameters. This would provide a new method to predict tobacco quality quickly and simply for tobacco field management and pertinence curing to tobacco operators. The innovation points of this research were as follows:First, remote sensing connected with GIS and GPS represented the development of modern agriculture. Recent years there were more and more researches on crop spectra, but little research were related to tobacco. Presently there were no qualitative and quantitative description of tobacco hyperspectra and no estimating models of many biophysical and biochemical indexes, mineral elements, so this restricted remote sensing techniques the to apply in tobacco growth inspection, yield estimation, quality inspection. This paper which analyzed tobacco hyperspectral characteristics from leaves to canopy, from traditional single direction to multi-angle observation and established the models of main biophysical and biochemical indexes, mineral elements made research systematism and provided more choices.Second, presently crop remote sensing applied mainly at certain growth period and growth conditions, so the models to estimate corresponding biochemical compositions by math analysis couldn't fit into other growth periods. With the changing of growth conditions, these models would be tested out. Therefore, we set up different treatments of tobacco types, flue-cured varieties and N P K levels in this experiment. The biophysical and biochemical indexes, mineral elements and spectral parameters were measured in different growth conditions and main growth periods, so the results which were obtained from all data removed the growth period and treatment influences effectively, and the models had stronger applicability.Third, we connected spectral parameters with flue-cured tobacco leaf maturity based on the former research which showed the close relationships between remote sensing parameters and many agricultural properties, especially chlorophyll. Then, we confirmed the red edge position as judging index of leaf maturity by correlation analysis and stepwise regression models. After analyzing the changing rules of spectral reflectance and the relationships between the red edge position and maturity, we made sure the red edge position of mature leaves and overripe leaves primarily, and tested the rationality by the estimating chlorophyll.Fourth, this research connected hyperspectral parameters and flue-cured leaf chemical quality, mineral elements, aroma compositions, and analyzed the relationships between them, and the models had good estimating effects. Although, the research about this was still in starting period, and the results were restricted by many factors, but it was a feasible way to predict tobacco quality for tobacco operators so it could provide references to pertinence managements. |
