Research On The Design Of Citrus Spectral Information Perception Parameterized Platform And Nondestructive Testing Of Sugar Content

In recent years,although near-infrared spectroscopy detection technology has been widely used in the field of fruit internal quality inspection,the development of near-infrared spectroscopy detection instruments for fruits in the field of instrumentation is still relatively backward compared with developed countries.The key core technology is firmly monopolized by foreign countries with high dependence on foreign countries.And the function of the existing instrument is too single to match the characteristics of fruits.As a result,the stability of non-destructive testing fruit quality by near infrared spectroscopy is poor.In view of the above problems,the aim of this paper is to improve the detection accuracy of soluble solids in citrus.Through the design of each unit of the parameterized platform module,the parameterized platform of citrus spectral information perception was developed to study the influence of light angle and spot size on the near-infrared spectral characteristics and detection accuracy of citrus with fruit pedicle facing downwards.In order to construct PLS prediction model of citrus sugar content,CARS algorithm was used to extract full spectrum characteristic wavelength under the best parameter matching of the platform,so as to accurately distinguish citrus with different sugar content.The main contents and conclusions of this study are as follows:1)Design and simulation analysis of key components of parametric platform.Based on the simplified model of the light source remote motion center mechanism,a light angle adjusting mechanism with adjustable spot size was designed.The adjustment range of the light angle was 0～60° and 1° was taken as the adjustment accuracy requirement of the light angle of a single light source.2)Testing mechanism performance of platform parameters.The minimum displacement deviation between the spot center and the target center of the four light sources at 60° was 7.5mm.The average illumination angles of No.2,No.3 and No.4 light sources were 0.2°,15.5°,30.8°,46° and 60.7° respectively at the reference illumination angles of No.1 light source(0°,15°,30°,45°and 60°),and the standard deviations were 0.15°,0.46°,0.60°,1.15°,1.33°,respectively.3)The transmitted light intensity of reference sphere was measured under different parameters.At 0～60° illumination angle,the average transmitted light intensity of the reference sphere with 35mm or 80mm diameter spot decreased with the increase of illumination angle,and the variation range of the light intensity was 0.754E+03～1.575E+04 and 1.447E+03～2.171E+04,respectively.At the same illumination angle,the average transmitted light of the reference sphere with an 80mm diameter illuminating spot was stronger than that of a 35mm diameter spot,and the range of light intensity difference was 0.596E+04～0.86E+04.4)The prediction model of citrus sugar content by diffuse transmission full spectrum under different parameters was established.The diffuse transmission spectrum of citrus fruit was collected with the fruit stalk facing down.Under the large spot,and when the illumination angle was 15°,the unprocessed full-spectrum partial least squares sugar content prediction model was the best,the correlation coefficient of prediction set was 0.7968,and the root mean square error of prediction set was 0.4697°Brix.Under the small light spot,and when the light angle was 30°,the unprocessed full spectrum partial least squares sugar content prediction model was the best,the correlation coefficient of prediction set was 0.8876,and the root mean square error of prediction set was 0.4560°Brix.5)Modeling of the characteristic wavelength at the optimum parameter matching.When the light spot was large and the illumination angle was 15°,and the full spectrum of the citrus with the stalk facing down was not processed.The citrus sugar content partial least squares modeling effect was the best after the characteristic wavelength was selected by the competitive adaptive reweighted sampling algorithm.The correlation coefficient and root mean square error of prediction set were 0.8619 and 0.3921°Brix,respectively.