| Sweet corn(Zea mays L.saccharate Sturt)is one of the most popular vegetables in countries such as the United States and Canada.Fresh sweet corn is rich in vitamins,amino acids and various trace elements needed by the human body.It is an important food for diet nutrition management in families due to its pleasant flavor and high nutritional value.In recent years,the popularity of sweet corn in the Chinese market has become higher and higher,and the popularity of market has led to the continuous development of the sweet corn planting industry.Hence,the planting scale has increased year by year.Guangdong Province is the main producing area of sweet corn in China,with an area of 220,000 hectares and a production of 3,102,000 tons per each year,the total planting area and annual output account for about half of the total in China,accounting for more than onetenth of the world’s total.However,sweet corn seed deteriorates more rapidly than other varieties of corn,which makes it difficult to manage the quality of sweet corn seeds during production,transportation and sowing processes.In order to overcome the disadvantages of the traditional methods for seed quality detection,such as time-consuming,high cost and destructive,it is of great significance to study on rapid and non-destructive quality detection technique for sweet corn seeds.In this study,commercial sweet corn seeds were used as experimental materials.The near-infrared spectroscopy(NIRS)technique and chemometrics were used to study the feasibility of heat-damaged detection,cultivar identification,vigor and viability detection of single kernel sweet corn seeds based on NIRS.Various prediction models for different detection purposes were trained based on the spectral data collected from single sweet corn seed,and proper results were obtained,which provided a reference for using NIRS to detect the quality of sweet corn seeds or other crop seeds with similar characteristics.The main research work of this paper is as follows:(1)To investigate the feasibility of detecting heat-damaged sweet corn seed based on single-kernel NIRS.The performance of three near-infrared spectrometers,namely,grating dispersion spectrometer,Fourier transform near-infrared(FT-NIR)spectrometer and linear variable filter spectrometer,for detecting heat-damaged sweet corn seed were comparatively analyzed.Partial least squares-discriminant analysis(PLS-DA)modeling algorithm was applied to train classification models for heat-damaged seeds detection.Since the scattering characteristic of heated-damaged seeds changed,baseline offset correction and multiplicative scatter correction(MSC)preprocessing methods were used to inspect the contribution of scattering characteristic on heat-damaged detection.Regression coefficients from PLS-DA models were used to identify the feature wavelengths for heat-damaged kernels distinguishing.The regression coefficient showed that the characteristic wavelength was related to the vibration absorption of starch and protein in the seed,which indicated that the molecular structure information of compounds in sweet corn seed changed after heating treatment.The performance of models basing on the spectral data collected from the embryo side and the endosperm side of sweet corn seed were compared.The models using the spectra of both sides could generate a high accuracy of 98.0%,indicating that the spectral data collected from the embryo side and the endosperm side of seeds contained information related to heat-damaged treatment.The performance of models which were trained from fullrange,short-wave range and long-wave range NIRS were compared,the results showed that both short-wave range and long-wave range NIRS contained the characteristic information of heat damage of sweet corn seeds.(2)Cultivar classification of single sweet corn seed using FT-NIR spectroscopy.FT-NIR technique was studied as a rapid and nondestructive technique to classify two cultivars of sweet corn seeds with high content similarity.The results of the composition determination showed that the relative content of soluble sugar,starch and cellulose in the two cultivars of seeds was less than 5%,and the relative content of lipid and protein was less than 0.5%.The characteristic wavelengths corresponding carbohydrate absorptions were identified,which caused spectral differences between the spectra of two cultivars.The effects of different modeling algorithms coupled with various preprocessing methods on classification accuracy were discussed.The multiplicative scattering correction(MSC)and standard normal variate(SNV)data preprocessing methods were found helpful for increasing the accuracy of Knearest neighbor(KNN),partial least squares discriminant analysis(PLS-DA)and support vector machine discriminant analysis(SVM-DA)models.The principal component analysis(PCA)was also used to inspect the structure of spectral data,as well as recognizing the spectral outliers.Each cultivar of samples had its own cluster center and two clusters were generally separated,which showed that the spectral data of the two cultivars was distinguishable.The highest recognition rates of the model based on full-range wavelength variable and feature wavelength variable were 100% and 99.19%,respectively,which demonstrated that the full-range model can be simplified without sacrificing too much accuracy by removing redundant wavelength variables.(3)Non-destructive detection of ageing and deteriorated sweet corn seeds based on FTNIR.The single-kernel FT-NIR spectroscopy with a wavelength range of 1000-2500 nm was used to detect the vigor of sweet corn seed after accelerated ageing treatment,and to detect the nonviable sweet corn seed treated by controlled deterioration operation.The standard germination test was performed to estimate the seed viability and vigor.The prediction models were trained for detection the vigor and viability of sweet corn seeds.The vigor grading of seeds at different ageing level could be classified correctly with the accuracies of73.6% and 64.0% for calibration and cross-validation,respectively.The principal component information of spectral matrix of deteriorated seeds and normal seeds was analyzed by performing PCA,and difference between the spectra of deteriorated seeds and normal seeds were found in the first six principal components.Based on the results of PCA,soft independent modeling of class analogy(SIMCA)algorithm was applied for establishing deteriorated seed detection model,the average accuracy of SIMCA models was 89.33%.The PLS-DA algorithm was also used for modeling,and the average accuracy increased to93.33%,which was higher than that of SIMCA models.The acceptable results indicated that applying FT-NIR technique for detecting deteriorated sweet corn seeds was feasible.Overall,basing on the experimental results of applying single-kernel NIRS technique for quality detection of sweet corn seeds,combined NIRS technique with preprocessing methods,multivariate data analysis and machine learning algorithms can successfully applied for detecting heat-damaged seeds,seed cultivars,ageing seeds and deteriorated seeds at a high accuracy level.Since the NIRS technique has the advantages of low-cost,highthroughput,and real-time detection,it shows great potential value in sweet corn seed industry for quality guarantee in production,transportation and sowing processes. |
PDF Full Text Request