Methods Research Of Corn Leaf Moisture Content And Nitrogen Detecting Based On Spectral Reflectance

The moisture content of corn leaves under different moisture treatments were predictedby the near-infrared (NIR) spectroscopy technique and hyperspectral imaging technique. Thereflectance spectral of corn leaves were acquired in the spectral range of300-2500nm byFieldSpec3spectrometer and300nm to1000nm or900nm to1700nm by a hyperspectralimaging system. The (MC) moisture content of the same samples was acquired by thetraditional oven drying method and nitrogen (N) of maize leaves was measured by automaticazotometer. Successive projections algorithm (SPA) method was applied to select the optimalwavelengths combined with multiple linear regression (MLR). The Soil Plant AnalysisDevelopment Unit (SPAD) index was investigated in order to select proper samples for Ntest.The main reseaych achievements were as follows:(1) The relations between near-infrared reflectance (300nm-2700nm) and MC werestudied. The Partial least square regression (PLSR) models were built by the full-bandwavelength. The results show the best model based on five pretreatment was standardnormalized variate (SNV) combined with PLS model. The prediction coefficients ofSNV-PLS model was increased from0.6978to0.7303and the root standard errors ofcalibration and prediction (RMSEP) was decreased from0.01909to0.01848.(2) Correlations were found between MC and hyperspectral imaging (900nm to1700nm).The results show the best model was based in raw full waveband. The coefficient was0.8650.The optimum wavelengths selected by SPA was1466,1330,1088,1589,1413,935,961nm.The coefficient of back-propagation artificial neural network (BP)-SPA was better thanPLS-SPA and MLR-SPA. The coefficient of the prediction set was0.8784. The coefficients ofthe three models were better than0.85with RMSEC and RMSEP less than0.01.(3) The PLSR model was applied to analyze the correlation between spectral data and Ncontent. The result of research shows that good correlations are found between N in maizeleaves and spectral information, the prediction R2is equal to0.8895with RMSEP of0.2915.So this work indicates that the rapid and non-destructive detect method of N in maize leavesbased on NIR spectroscopy technique is high accuracy and practical value.(4) Correlations were found between N of the corn leaves and hyperspectral imaging (400nm to1000nm). The results show the Baseline-PLS model based on full waveband wasbetter than seven pretreatment. The coefficient was0.8796. The optimum wavelengthsselected by SPA were723.28、767.12、442.22、434.91、437.35nm. The coefficient ofback-propagation artificial neural network baseline-SPA-BP was better thanbaseline-SPA-PLS and MLR-SPA. The coefficient of the prediction set was0.8717. What’smore, the coefficients of the three models were better than0.82.(5) Correletion between N and SPAD is y=0.042x+1.033，The coefficient was0.8567。The results show the De-trending-PLS model based on raw full waveband was better thanother seven pretreatment.29optimum wavelengths were selected by SPA, andDetrending-SPA-BPNN is best，he coefficient of the prediction set was0.8784.