Detecting Moisture Content And Vigor Of Maize Seeds Based On Spectroscopy Technology

Maize is one of the most important food crops in China.How to increase maize output has always been an important issue in China’s agricultural development,and the quality of maize seeds is a prerequisite for improving maize yield and quality.Among them,seed moisture content and seed vigor are important indicators for evaluating seed quality.In order to provide a method for non-destructive and rapid detection of maize seed moisture content and vigor,with a view to achieving industrialized detection,visible/near-infrared（Vis/NIR）and near-infrared（NIR）hyperspectral imaging technology（375.18～1017.88 nm,865.11～1711.71 nm）and spectroscopy technology（650～1100 nm,900～1700 nm）were used to collect spectra of maize seeds of various varieties.SG smoothing,standard normal variable correction（SNV）,multiplicative scatter correction（MSC）,first derivative(1^stDer),method was used to divide the sample set.Uninformative variable elimination（UVE）and competitive adaptive reweighted sampling algorithm（CARS）were used to extract characteristic wavelengths,and partial least squares regression（PLSR）and support vector machine（SVM）were used to establish quantitative prediction models of moisture content and vitality（germination rate,germination index and vitality index）of different maize seeds,respectively.The main research contents and conclusions of this article are as follows:（1）The method of predicting the moisture content of maize seeds based on Vis/NIR and NIR hyperspectral imaging technology was studied,and prediction models were established.Based on the hyperspectral images of the seed embryo and endosperm surfaces of four maize varieties,the effects of average spectral extraction regions（centroid area and whole seed region）and different spectral preprocessing methods on the model accuracy were studied.The feature wavelengths extracted by UVE and full spectra were used to build PLSR prediction models.The results showed that the models built by the average spectra extracted from the centroid region were superior to the spectrum extracted from the whole seed region;the SG smoothing method was superior to other pretreatment methods;the PLSR models based on the NIR spectra were better than those based on the Vis/NIR spectra;the prediction model based on a single variety（RMSEP≤2.001%）was better than the model based on a mixture of four varieties（RMSEP≤2.116%）.（2）Researched the method of predicting the moisture content of maize seeds based on Vis/NIR and NIR spectroscopy.Based on the Vis/NIR spectra of seed embryo and endosperm surfaces of six varieties in the range of 650～1100 nm and NIR spectra in the range of 900～1700 nm,the effects of different spectral pretreatment methods on modeling accuracy were studied.PLSR models based on the full spectra and the characteristic wavelengths extracted by CARS and UVE were built.The results showed that S-G smoothing was superior to other spectral pre-processing methods.Most PLSR models built using the characteristic wavelengths selected by CARS had better performance.Compared with Vis/NIR spectroscopy,the models built based on NIR spectroscopy could better predict the moisture content of maize seeds.Similarly,the model based on a single variety（RMSEP≤1.702%）can better predict the moisture content of maize seeds than a mixture of six varieties（RMSEP≤1.910%）.（3）Explored the method of predicting the germination rate,germination index and vigor index of maize varieties based on Vis/NIR and NIR spectroscopy.Based on the Vis/NIR（650～1100 nm）and NIR spectra（900～1700 nm）of the seed embryo and endosperm surfaces of three maize varieties,the influence of the spectral pretreatment method on the prediction accuracy was analyzed,and the UVE and CARS were used to select characteristic wavelengths,PLSR and SVM prediction models based on characteristic wavelengths and full spectra to predict germination rate,germination index and vigor index were established.The results showed that for germination rate and vitality best pretreatment method.In most cases,using CARS to select characteristic wavelengths had better model prediction performance.For the germination rate,the model based on PLSR had a better prediction effect,while for the germination index and vigor index,the SVM model had a better prediction effect.Compared with Vis/NIR spectroscopy,NIR spectroscopy had better prediction performance in predicting germination rate,germination index and vigor index.Among the three vigor indicators,the effect of predicting the vigor index was the best,and the R_P could reach 0.924.The results of this study indicated that hyperspectral imaging technology and spectroscopy technology could predict the moisture content and vigor（germination rate,germination index and vigor index）of many varieties of maize seeds,and NIR spectra had better prediction performance than Vis/NIR spectra.