Study On Quantitative Model Of Cotton Growth Information Based On Hyperspectral Data

[Research object] This essay based on different treatments of cotton in XinJiang, utlizi- ng hyperspectral remote sensing , cotton canopy have been carried on multi-time monitoring, some growth information at key growing stages has been extractd. Based on spectral deriv- ation technology and some analysis-compute methods, aiming at the correlation between hyperspectral data and some structural status parameters, such as leaf aera index(LAI), biomass,vegetation frection(fv) and so on. Some suitable hyperspectral parameters have been chosen and several cotton growth information quantitative models will be established.[Research methods] In field, main cultivated varieties XinLuzao-13 and XinLuzao-19 have been chosen North in XinJiang.Using 19 hyperspectral variables and 18 vegetation indices, respectively established the relationship between the 37 variables and LAI,biomass, fv, then, simulated cotton growth information regression models. Through validation and precision evaluation, several estimation models have been primary designated.[Innovation] The digital photoes of cotton canopy will be inset MatLab6.5 ,the fv value of whole growth stages would be got.In this essay, some important vegetation indices were used. There were three styles perpendicular vegetation index(PVI-1³), many wave band composed NDVI(NDVI-1⁷), SAVI, as well as RNVI. The estimation models of cotton fv were new.[Research content and results] were as following :(1) In all growing stages, the cotton LAI changed regularitily indicated that LAI arrived peak value at blooming-boll stage under different varietyies density and planting styles treatment. The LAI acted S curve trend. The spectrum and LAI took on synchronal characteristic, also indicated the variation of LAI whole stages. According to the positive relationship between LAI and yield, for the sake of reaching ideal LAI and higher yield, the cultivated type of equidistant line and small-three film were better than big-three film or machinery picking types. The planting density of 22.5×10⁴ plant·hm^-1 was fit. All the determination coefficients of XLZ-13 were higher than whose of XLZ-19. The data of XLZ-13 act as the experimental sample, the data of XLZ-19 act as the test sample, the result of analysis indicated that the simulation model′s precision was during 77%⁸5%. Especially, two exponential functions were the best, which hyperspectral variable were NDVI-5,the RMSE was very small, the precision was above 85%. Therefore, the exponential function was the optimum estimation model for cotton LAI in all growth stages.(2)The increase of cotton′s aboveground dry matter accumulation behaved a trend of S curve. Form blooming to full-boll stage was the key stage of dry matter acumulation , the peak stage was the early boll-opening stage. Under four planting styles, The correlation analysis between 37 spectral parameter and six agricultural parameter, (there were fresh or dry leaf weight, bud-bloom-boll weight,above ground fresh or dry biomass, for short:LFW, LDW, BFW, BDW, AFM, BFM) which regression model′s type were non-linear. The data of XLZ-13 was experiment sample, the data of XLZ-19 was examination sample, the three variables of SDr/SDb,(SDr-SDb)/(SDr+SDb) and SDr/SDy through testing, three models have been chosen, which relative error was very low, the correlation between the estimated value and the measured value was higher. Thus, the three parabolar pattern models, for example, Y=-0.5719 SDr/SDb²+17.993 SDr/SDb -111.19,Y=31.976(SDr-SDb)/(SDr+SDb) 16.466,Y=-0.0035 SDr/SDy²-0.2484 SDr/SDy -2.9219 acted as the best models, which estimated LDW, AFM,BFW and ADM utlizing hyperspectral data in all growth stages.(3) The cotton vegetation fraction correspond cotton′s growing regular pattern, can respond cotton growth status. Under four planting styles, the two varieties′fv value arrived the highest level at full-bolling stage. The type of equidistant line has a higher fv. Using 16 spectral parameters act variables, these non-linear models have higher correlation coefficients. Thus, Rg,NDVI-1⁷,PVI-3 ,et al were the most significant indices. The data of XLZ-19 was experiment sample, the data of XLZ-13 was examination sample, using SDr/SDb as variable to simulate cotton′s fv, the determination coefficient of the parabolar pattern models was the biggest, the models′precision aboved 0.9.So this model was the best.