| Apple is an important economic fruit in China.High-throughput sorting of postharvest quality is helpful to improve the value of the apple industry and achieve the goal of apple quality and internationalization.With the improvement in living standards,consumers pay more and more attention to the quality of apples.Therefore,it is very important to use nondestructive testing technology to realize the rapid detection of apple internal quality and diseases.In this study,the ’Fuji’ apple was used as the research sample.Based on Vis-NIR spectroscopy technology,an online detection module for moldy apple core and soluble solids content(SSC)was designed.The online detection model of apple SSC was established by online apple spectral data and physical and chemical indicators.The influence of apples’ external and internal quality on the discriminant model of moldy core was explored.And the influence of diameter was corrected based on the attenuation coefficient of light inside an apple,and the influence of internal soluble solids on the moldy core model was corrected by intermediate data fusion and variational mode decomposition algorithm.Finally,the prediction accuracy of the online detection model is verified.The main research contents and conclusions are as follows:(1)The online detection module of apple moldy core and SSC was designed.By comparing different spectral acquisition methods,it was found that the diffuse transmission spectrum acquisition method composed of two 100 W halogen lamps had a better effect on predicting apple SSC and moldy heart disease.Single factor test and central response surface test were carried out on different influencing factors of the designed online detection module to optimize the parameters of the online detection module.It is found that when the integration time is 91 ms,the distance from the light source to the fiber is 169 mm,the light source voltage is 12 V,and the line selection speed is 0.14 m/s,the stability of the online detection module is high.Based on the response surface analysis of the interaction between various influencing factors,it is found that the order of the influence of each factor on the stability of the detection module is the speed of the line selection,the voltage of the light source,the integration time and the distance from the light source to the fiber.The influence of each influencing factor on the stability of the detection module is not a simple linear relationship.Based on the optimal acquisition method and operating parameters,an online spectral measurement system was built,and the performance of the spectral online detection module was verified by recollecting the spectral data of PTFE spheres.The theoretical value of the spectral signal-to-noise ratio was 37.33,and the actual value was 36.53,indicating that the predicted value and the actual value fit well,and the stability of the collected spectrum of the online detection module was verified.(2)The prediction model of online apple SSC was established.The online prediction model of apple SSC based on the full spectrum was established by using linear regression,nonlinear regression,ensemble regression,and neural network regression algorithms.By determining the important features of the machine learning model,the overlap rate of important features under different data sets is calculated,and the reasons for the differences in the prediction effects of different models are explored.It is found that the performance of the prediction model based on the partial least squares regression algorithm is better than other machine learning models.The Rp is 0.92 and the RMSEp is 0.50°Brix.The overlap rate of important features extracted by the partial least squares regression algorithm is high,which can better explain the high robustness of the linear regression algorithm.By recollecting samples,the prediction accuracy of the constructed Apple SSC online detection model was verified.It was found that the SSC prediction model based on the partial least squares regression algorithm was the best,with Rp of 0.88 and RMSEp of 0.59°Brix,which was consistent with the previous results.(3)The effect of apple’s internal and external quality on the online discriminant model of moldy apple core was explored.Using the attenuation coefficient of light inside the apple to correct the diffuse transmission spectrum,an online prediction model of the moldy apple core based on the diameter correction spectrum was established.It was found that this method can effectively reduce the error of apple diameter in the prediction process.The moldy apple core discrimination result is 3.71% higher than the model based on the original spectrum,which improves the online discrimination accuracy of a moldy apple core.The distribution of SSC in healthy apples and moldy apples was analyzed.It was found that the SSC of moldy apples was generally lower than that of healthy apples,indicating that there was an effect between apple SSC and moldy apples.The online discriminant model of moldy apple core considering apple SSC was constructed by using the intermediate data fusion method and variational mode decomposition algorithm.The intermediate data fusion method was to fuse the relevant characteristics of moldy core and SSC,and the variational mode decomposition algorithm weakened the spectral components related to SSC.The results showed that both methods improved the online discriminant accuracy of moldy core,and the highest discriminant accuracy of moldy core was 94.44%.The discriminant accuracy of the online detection model to moldy apple core was verified.It was found that in the independent sample set,the moldy core using the intermediate data fusion method and the variational mode decomposition algorithm still had good prediction performance.Among them,the model based on the variational mode decomposition algorithm had the highest discriminant accuracy,and the discriminant accuracy was 88.33%.The verification results show that the online detection model has high stability and can be used for online detection of apple moldy core. |
PDF Full Text Request