Spectral Doagnostics Methods On Growth Information Diagnostics Of Gannan Navel Orange Leaves

With the development of precision agriculture, a rapid and non-destructive diagnosticneed was proposed on growth information of Gannan navel orange leaves. Chlorophyll, waterand nitrogen were important nutritional information in the course of navel orange leaf growth.There were inevitably problem, such as complex detection, damage samples, time-consumingand so on in traditional measurements of leaf growth information. Spectral diagnostics hasfast, non-destructive and efficient advantages.In this study, building hyperspectral imagingdevice and existing near-infrared Fourier spectrometer were used to detect chlorophyll, waterand nitrogen in Gannan navel orange leaf. Spectra pretreatment and band selection methodwere reseached mainly for optimal model of three indicators, which provided the basis for thespectral diagnostic in growth information of navel orange leaves. The main work andconclusions of this paper were as follows:1) The hyperspectral imaging device was set up. The structure and build process ofhyperspectral imaging systems were introduced from acquisition module, imaging modules,optical modules, lighting, camera obscura and software, which lay the foundation for thegrowth of navel orange leaves.2) Genetic algorithm (GA), successive projections algorithm (SPA) and competitiveadaptive reweighted sampling (CARS) were discussed the pros and cons for extractingvariables of apectral data. The filtering variables were modeled by partial least squares (PLS),All of GA, SPA and CARS can effectively extract characteristic variables from the spectraldata, which greatly simplifies the model. Howevwe, SPA is not only stability but alsoextracting least characteristic variables, but relatively weak in terms of accuracy. GA andCARS were slightly less stability, but a higher accuracy of the model created based on theextracted variable.3) Hyperspectral imaging device had a better result in detecting chlorophyll content ofnavel orange leaf. Measured chlorophyll region was chose as the region of interest (ROI)from the acquisited hyperspectral images, whose average spectral were prosessed andanalysised, the PLS model with extracting variables optimum effect. Prediction modelcorrelation coefficient (R p) and standard deviation (RMSEP) were0.96,2.14, which can beused as the best model of hyperspectral imaging technique to detect the chlorophyll of navelorange leaf.4) The Fourier near-infrared (FT-NIR) spectrometer had a better result in detecting watercontent of navel orange leaf. First, the leaves FT-NIR spectroscopies were preprocessed by multiplicative scatter correction (MSC). After selected variables, models were created.GA-PLS model achieved optimum results,R pand RMSEP were0.96,1.75, which could beas the best model of FT-NIR to detect the water of navel orange leaf.5) Hyperspectral imaging device had a better result in detecting nitrogen content of navelorange leaf. Entire leaf area was chose as the region of interest (ROI) from the acquisitedhyperspectral images, the ROI average spectral was processed and analysised, found that theeffect of CARS-PLS model is excellent,R pand RMSEP were0.82,0.39. Hoever, thenitrogen model results was insufficient compared to chlorophyll and water.6) Fusion spectral diagnosis the growth information of navel orange leaf. The data ofhyperspectral imaging and FT-NIR were fused. With respect to the separate model ofhyperspectral and FT-NIR, the accuracy of fusion model was little difference, the principalcomponent declined.