Study On Rapid Detection Of Physiological Indexes Of Muskmelon At Different Stages Under Light Environment Regulation

In this study,melon "M5147" and "M5346" were used as test materials,and different treatments(light ratios and light intensities)were set up to select a suitable light environment for the plant factory crop growth and the model for accurate and rapid monitoring of plant growth.By measuring growth indexes such as plant height,stem diameter,leaf number and node number,root morphology index,leaf chlorophyll content,chlorophyll fluorescence parameters and antioxidant enzyme activity,the differences in the response of melon in plant factory to different light ratio and light intensity were explored.The prediction models of chlorophyll content,maximum photochemical efficiency(Fv/Fm)and superoxide dismutase(SOD)of melon were established by hyperspectral imaging(HSI)techniques(visible near infrared HSI,fluorescence HSI,microscopic HSI).The original spectrum was preprocessed and optimized by six preprocessing methods.The feature wavelengths were extracted by competitive adaptive weighting algorithm(CARS),successive projections algorithm(SPA),uninformed variable elimination(UVE)and genetic algorithm-combined partial least squares(GAPLS).Partial least squares regression(PLSR),least squares support vector machine(LSSVM)and convolution neural network(CNN)quantitative prediction models were established based on the optimized characteristic wavelength.The results of this study provide data support and theoretical basis for the screening of light ratio and light intensity of other crops in plant factories,and provide technical support for on-line monitoring of crop growth in plant factories.The main results are as follows:(1)T4(Red:Blue:White:UV=7:3:5:1)was preferred for later light intensity test,and T3(480μmol/(m2·s))was preferred for post photoperiod test.(2)Three methods,Random Sampling(RS)、Kennard-Stone(KS)and sample set partitioning based on joint x-y distance(SPXY)were used to select the calibration and prediction sets from 216 melon leaves.Six preprocessing methods were used to optimize the original spectral data.The Baseline、Normalize、Orthogonal signal correction(OSC)and Standardized normal variate(SNV)preprocessing methods were optimized to establish the PLSR model for chlorophyll content,and the OSC and Normalize preprocessing methods were optimized to establish Fv/Fm and SOD PLSR models.(3)CARS,SPA,UVE,and GAPLS methods were applied to extract characteristic wavelengths and construct PLSR models.PLSR,LSSVM,and CNN models were established based on the optimal characteristic wavelengths.Through comparison and analysis,the CNN method was found to be more effective in establishing models for chlorophyll content(Rc=0.883,Rp=0.676),Fv/Fm(Rc=0.931,Rp=0.854),and SOD(Rc=0.914,Rp=0.699).(4)The correlation coefficients Rc and Rp of the CNN model for chlorophyll content based on HSI and fluorescence HSI techniques were higher than those of microspectral HSI,with values of 0.949 and 0.798,respectively.The correlation coefficient R of the CNN model for Fv/Fm based on fluorescence HSI technology was higher than that of both HSI and microscopic HSI technologies,with the value of 0.915.The correlation coefficient R of the CNN model for SOD based on fluorescence HSI and microscopic HSI technology was higher than that of HSI technology,with the value of 0.867.Overall,the performance of the CNN prediction model for physiological indicators of melon based on fluorescence HSI technology was the best.