| Nondestructive testing is of great significance for the rapid development of fruit quality and industry.The spectral data obtained by hyperspectral imaging system is very large and the equipment cost is high.Machine vision technology has certain practicability for fruits with not obvious surface curvature,and can accurately obtain fruit surface images.For most fruits with large surface curvature,it is difficult to quickly obtain all and true images of the surface.The visible/near infrared spectroscopy technology has the advantages of fast,convenient and non-destructive,and has been gradually used in the detection of agricultural products,avoiding the disadvantages of repeated destructive testing of samples,and can achieve rapid and accurate identification of fruit quality.Therefore,it is of great significance to explore the visible/near infrared spectroscopy technology for comprehensive and systematic monitoring of the quality of table grapes.In this study,the spectrum of grapes was collected by handheld optical fiber spectrometer,and the maturity indexes of grapes were systematically and comprehensively associated for modeling and analysis,reflecting the maturity quality of grapes and providing theoretical and practical basis for online monitoring of grape quality.The main research results are as follows:1.The samples were divided into four ripening stages,i.e.,ripening stage Ⅰ,ripening stage Ⅱ,ripening stage Ⅲ and ripening stage Ⅳ,according to the pattern of changes in soluble solids,total acid,pH,total phenols of grape skins,tannins of grape skins,total phenols of grape seeds,tannins of grape seeds,color,texture,grain size and weight.The variation patterns were analyzed.Soluble solids,soluble solids/total acid and pH increased with the progression of ripening stages,total acid values decreased,phenolics and tannins decreased,a*and c*values increased,L*and b*decreased significantly,h*values decreased significantly,and c*values reached the highest value in ripening stage Ⅳ.Hardness showed an overall decreasing trend,with a sharp decrease from ripening stage Ⅰ to ripening stage Ⅲ,and hardness in ripening stage Ⅲ and The hardness of grapes at ripening stage Ⅲ and Ⅳ did not change significantly,consistent with the decreasing trend of chewiness.Fruit elasticity differed significantly between ripening stage Ⅰ and ripening stage Ⅱ,and between ripening stage Ⅲ and ripening stage Ⅳ,and fruit grain size and single grain weight increased slowly.2.Eleven grape tasting panel members were organized to taste grapes at ripening stage Ⅰ,ripening stage Ⅱ,ripening stage Ⅲ and ripening stage Ⅳ for sensory scoring,which showed that grapes at ripening stage Ⅲ had the highest scores and the overall level of grapes at this stage was better,followed by ripening stages Ⅱ and Ⅳ,and ripening stage Ⅰ had the lowest total tasting scores,and there were significant differences between ripening stage Ⅲ and ripening stages Ⅱ and Ⅳ and ripening stage Ⅰ.3.Six spectral models for predicting grape ripeness were developed based on full wavelength and eigenwave lengths using partial least squares algorithms by combining six preprocessing methods,namely standard normal transform,multiple scattering correction,first-order derivative,second-order derivative,S-G convolution smoothing,and S-G convolution smoothing+first-order derivative.Soluble solids were modeled based on the full wavelength combined with the spectral preprocessing method of first-order derivatives,and the coefficients of determination for the correction set(R~2Cal)and the prediction set(R~2Pre)were 0.97 and 0.93,respectively,and the root mean square error of the correction set(RMSEC)and the root mean square error of the prediction set(RMSEP)were 0.62 mg/100g and 1.27mg/100g,respectively,with an RPD of 4.09;the coefficient of determination of the correction set(R~2Cal)and the prediction set(R~2Pre)based on the full-wavelength modeling of total acid,combined with the spectral preprocessing method of first-order derivation,was 0.97 and 0.94,respectively,and the root mean square error of the correction set(RMSEC)and the root mean square error of the prediction set(RMSEP)were 0.88 mg/100g and 1.96 mg/100g,respectively,with RPD was 4.55.4.The determination coefficients of the correction set(R~2Cal)and the prediction set(R~2Pre)were 0.90 and 0.82,respectively,based on the full-wavelength modeling and the spectral preprocessing method of S-G convolution smoothing+first-order derivative.Root mean square error(RMSEC)of correction set and root mean square error(RMSEP)of prediction set were 4.70mg/100g and 6.29mg/100g,respectively,and RPD was 2.71.The determination coefficients of the correction set(R~2Cal)and prediction set(R~2Pre)were 0.96and 0.95,respectively,based on full-wavelength modeling and combined with the spectral preprocessing method of multiple scattering correction.Root mean square error(RMSEC)of correction set and root mean square error(RMSEP)of prediction set were 2.50mg/100g and2.52mg/100g,respectively,with RPD of 5.11.The determination coefficients of the correction set(R~2Cal)and prediction set(R~2Pre)were 0.81 and 0.73,respectively,based on full-wavelength modeling and spectral preprocessing method with second-order derivative.The root mean square error(RMSEC)of the correction set and the root mean square error(RMSEP)of the prediction set were 6.45mg/100g and 8.58mg/100g respectively,and the RPD was 1.59.The determination coefficients of the correction set(R~2Cal)and prediction set(R~2Pre)were0.85 and 0.84,respectively,based on the characteristic wavelength modeling and the spectral preprocessing method of second-order differentiation.The root mean square error of the correction set(RMSEC)and the root mean square error of the prediction set(RMSEP)were2.95mg/100g and 3.23mg/100g,respectively,and the RPD was 2.96.The study showed that the physical and chemical indexes of grape at four maturity stages changed significantly,and the evaluation results were significantly different.The prediction set determination coefficients of the model established by combining the visible/near infrared spectrum with the six main physical and chemical indexes were mostly between 0.80 and 0.95,and the model was relatively ideal.This model provides practical guidance for nondestructive testing of grape quality and maturity and theoretical basis for further on-line monitoring of fruit quality. |
PDF Full Text Request