| Traditional way of fruit sorting relies mainly on the human eye of fruit weight,size and appearance quality such as sorting,this method is difficult to distinguish its internal quality,and near infrared dynamic online non-destructive testing equipment can realize the thin skin and the skin of fruit internal quality sorting,and forecast model accuracy is close to one of the core technology of infrared dynamic online sorting equipment,High-precision model is an important guarantee for fruit sorting.In order to improve the model accuracy,this paper took thin-skinned fruit(Gansu Red Fuji apple)and medium-skinned fruit(Guang Xi Fertile orange)as the research objects,and adopted near-infrared diffuse transmission method to collect spectra.In circular light 1000 w(10 halogen lamp),fruit acquisition speed per second for 5spectral acquisition under working conditions,combined with stoichiometry to find out the two typical fruit under different sizes of the optimal parameters,and the optimal parameters of the model to optimize the model,in order to realize the rapid and high precision of the typical fruit internal quality inspection.The main research results and conclusions are as follows:(1)Thin-skinned fruit(Gansu Red Fuji apple)was taken as the research object in this paper.Apple models with different sizes(70-80 mm,80-90 mm,90-100 mm)were established by using NIR nondestructive dynamic online testing device under three integrating time(75 ms,100 ms,125 ms)and three directions,and the sugar content of the models was evaluated.Firstly,the influence of size on the model accuracy was analyzed,and it was found that the maximum size could have 81.1%influence on the sugar content model.Secondly,the combination of apple size and integration time was studied under different directions.Under the combination of specific directions and integration conditions,the effect of sugar content model was 98.4%.In the range of 610-1025 nm,710-980 nm,750-1025 nm and so on,multiple scattering correction(MSC),normalization(Normalize)and Gaussian filtering smoothing(GFS)were used to preprocess the specific direction of the spectrum.Among them,the normalized method has the best effect at 710-980 nm,and its model parameters areR_p~2=0.869,R_C~2=0.958,RMSEC=1.240E-06 and RMSEP=2.220E-06.Finally,competitive adaptive weighted sampling(CARS)and uninformative variable elimination(UVE)were used to screen the characteristic wavelengths of the selected normalized effective wavelengths at 710-980 nm.It was found that CARS-PLS had the best effect,and the number of variables decreased from 360 to 34,R_C~2 increasing from 0.958 to 0.971.R_p~2 From 0.869 to0.948.This study provides a good reference for the optimization of apple sugar content model with different sizes.(2)In this paper,the medium-skinned fruit(Guangxi Fertile orange)is taken as the research object.The sugar content models of different sizes,directions and integration time were established to find the optimal parameters of different sizes of tangan,and the model was optimized to realize the high-precision detection of internal quality of tangan.Firstly,the spectra of apples in three different sizes(70-75 mm,75-80 mm,80-85 mm)were collected at three integrating times(75 ms,100 ms,125 ms)and in three directions by using the NIR nondestructive dynamic online detection device.Then,81 sugar content models of tangan were established by USING PLS algorithm,and it was found that the integration time combination of small,medium and large fruits under the direction of the third quadrant and 75ms-100 ms-125 ms had better effect.The model parameters wereR_p~2=0.804,R_C~2=0.945,RMSEC=0.317 and RMSEP=0.605.Finally,a variety of pretreatment algorithms were used to process the spectra in the spectral range of 710 nm-980 nm,and it was found that normalized pretreatment achieved the best effect.After a variety of band screening algorithms were adopted to select the spectrum,it was found that CARS-PLS modeling had the best effect,and its variables were reduced from 360 to 24 by 345,R_C~2 was slightly reduced from 0.927 to0.901,R_p~2 was increased from 0.869 to 0.878,and the number of modeling variables under this model was reduced to a certain extent.The modeling level effect has been improved to some extent.Tangan with a small size span is not the same as apple. |
PDF Full Text Request