| Chlorophyll fluorescence(ChlF)—an important tool for revealing the photosynthesis process of photosystemⅡ(PSⅡ)—has been widely accepted and investigated by physiologists due to its non-invasive measurement.The maximum quantum efficiency of PSⅡ(F_v/F_m)could indicate plant’s growth status and response to the environment.The conventional measurement of F_v/F_mhas two main disadvantages:high expenses of the chlorophyll fluorometer,long-time measurement,which limits its application in agricultural production.To obtain ChlF information rapidly from a large scale,researchers explored the correlation between ChlF parameter and vegetation indices(VIs),and their results demonstrated the possibility of evaluating ChlF parameter by reflectance light signal of few wavelengths.However,restricted by the data dimension and fitting methods,the accuracy of those research is limited.In addition,the expense of spectrometer is relatively high.Facing these challenges,this research extracted key wavelengths for evaluating F_v/F_m from visible-near infrared(Vis-NIR)spectrum of plant leaves,constructed a portable detector for F_v/F_m with a core of light emitting diodes(LEDs)and photodiode,established F_v/F_m calibration models for the detector,developed cloud server for model running and data storage,and an android application for model selection and result display.The main research contents and results are as follows:(1)Extraction of key wavelengths for evaluating F_v/F_m.Considering the influence of plant species and stress types on the relationship between ChlF parameter and spectrum,eggplants exposed to high-light stress and cucumber plants exposed to low-temperature stress were obtained as experimental materials.F_v/F_m and reflectance spectrums of their leaves were measured.For both plants,their Vis-NIR spectrums decreased with the increase of F_v/F_m.To obtain few wavelengths which are important for evaluating F_v/F_m,3 pretreatment methods were used for spectral data preprocessing,5 variable screening algorithms were used to obtain important variables and obtain characteristic wavelength combinations,Partial least squares regression(PLSR)was used to establish F_v/F_m regression model for variable evaluation.The accuracy of the regression models and the variable regression coefficients of the optimal model were analyzed.6 characteristic wavelengths for F_v/F_m were confirmed to be 400nm,525nm,600nm,650nm,730nm,740nm.(2)Hardware development of the detector.The hardware mainly consists of 5components:control unit,light source unit,detection unit,input/output unit and communication unit.The control unit has a micro-controller of STM32F103,reads the detected data and controls other units.The light source unit includes an LED ring array and driving circuit,driving the light source to emit detection light in a constant current mode.The detection unit includes photoelectric conversion circuit and signal conditioning circuit to achieve signal conversion,noise reduction and amplification.Input/output unit uses OLED display and button circuit to achieve user-device interaction;communication unit uploads the detection data to a Cloud Server through the NB-Io T protocol.The hardware driver program was achieved by C language for function realization.The reflected light transmission path of the blade surface was analyzed and the reflected light detection probe was constructed accordingly.The shell structure was designed to meet the light weight property of the detector.(3)Establishment of F_v/F_m prediction models for the detector.Light and temperature stress gradient experiments were set up for eggplants and cucumber plants.The F_v/F_mand reflected light of the leave samples were measured by chlorophyll fluorometer and the detector,respectively.2 F_v/F_m prediction models were established for eggplant under high-light stress.Support vector regression algorithm was used modeling,and grid search was adapted for parameter optimization.The modeling performance of propagation(BP)neural network,radial basis function(RBF)neural network,extreme learning machine(ELM)and SVR algorithms was compared.The results indicated that the SVR algorithm showed the best performance with average determination coefficients(R~2)of 0.8114,0.7984,and average root mean square error(RMSE)of 0.0169,0.0178 for eggplant and cucumber,respectively.(4)Detector software development and performance testing.According to the calculating resource required for running the SVR algorithm and the potential need for model updating,a cloud server was developed based on the Alibaba Cloud computing platform and configured with My SQL database and Python environment for data storage.Also,an android application was developed achieve model selection,detection result display and history query.Experiments were designed to test the dark current signal,reflected light detection effect and F_v/F_m detection accuracy.The results showed that the dark current signal was low and the disturbance was small;The deviation of the repeated detection data of the detector was small,and the data can effectively reflect the difference of leaf samples with different F_v/F_m;The F_v/F_m of eggplants and cucumber plants was measured by the detector and a chlorophyll fluorometer,showing R~2of 0.7327 and 0.7489,RMSE of 0.0186、0.0228,respectively. |
PDF Full Text Request