On-line Detection Of Citrus Sugar Content Using Near-infrared Spectroscopy Based On Different Postures

Citrus is one of the most important fruit varieties in China.Visible/near-infrared spectroscopy analysis technology is widely used for quality testing of citrus due to its advantages of non-destructive,fast,low-cost,simple operation,and green environmental protection.The non-destructive testing and grading equipment for fruit near-infrared spectroscopy on the market is also gradually being applied.However,as a complex mixed natural agricultural product,differences in sample related factors such as fruit color and size,as well as differences in sample unrelated factors such as the posture and environmental factors of citrus placement during postharvest spectral testing,pose difficulties and challenges for online non-destructive testing of citrus.This study focuses on Shimen tangerine and focuses on the impact of citrus detection posture factors.It conducts an online detection of citrus sugar content based on a double cone roller fruit cup conveying system.The main research content and conclusions are as follows:(1)A citrus internal quality sorting production line based on a double cone roller conveyor system has been designed and constructed.Firstly,clarify the design composition and workflow of the entire sorting production line system.Secondly,an online sugar content detection module based on a fully transparent optical path was designed and built,and the integration and triggering methods of online spectral collection were compared.The spectral delay collection method triggered by citrus photoelectric sensors was clearly adopted,and the entire sugar content online detection hierarchical control system was designed.Finally,the static and dynamic spectral collection performance of the entire system was tested.The relative standard deviation of the sample spectra collected multiple times in the commonly used band range of 5001000nm for citrus sugar content detection was not higher than 0.05,indicating good collection stability.(2)Analyzed the effects of different spectral acquisition postures and locations on the transmission spectra of citrus.Six representative detection postures(G1-G6)were determined,and spectral variation between different postures was analyzed using spectral difference spectrum.The relative standard deviation was used to compare and evaluate the changes in the positions of peaks and valleys,and the single factor analysis of variance was used to analyze the changes in light intensity values at the peaks and valleys.It was concluded that attitude has little effect on the positions of peaks and valleys,and there is a significant difference in the magnitude of light intensity values at the peaks and valleys under the six postures;At the same time,error bands and relative standard deviations of spectral area were used to analyze the impact of acquisition positions on spectral stability under various postures,and it was found that the G1 attitude is a relatively stable spectral acquisition attitude.(3)The influence of different postures on online detection of citrus sugar content models was studied.Firstly,a comparative study was conducted on the effectiveness of partial least squares(PLS)models for citrus sugar content established under different spectral data types and band selection.The optimal data type for establishing sugar content models based on original light intensity spectra and 550-950nm bands was identified.Secondly,after removing abnormal samples and comparing different preprocessing methods,a single pose sugar PLS model and a global model fused with six poses were established for cross validation.It was found that the six single pose models only had certain predictive ability for the spectral data of their own pose,while the global pose model had good stable predictive ability for all six poses.Subsequently,the global pose model was optimized for feature variable selection,And conduct external verification under random attitude on the machine.Finally,based on the characteristics of online citrus transportation,a multi pose combination model integrating eight high-frequency citrus poses was established,and the root mean square error of the prediction was reduced to 0.779° Brix through external validation on the computer.(4)A two-step prediction model for citrus sugar content based on posture classification has been proposed for online detection to improve the accuracy of sugar content detection under random postures.Using multi posture citrus spectral data,we tried to establish a discriminant analysis model for citrus posture,compared and determined the partial least squares discriminant analysis(PLSDA)algorithm,and carried out pretreatment and feature variable screening optimization.The best discriminative model was Mean Centering-PLSDA,and the classification accuracy of the test set reached 95%.The root mean square error of prediction(RMSEP)using a two-step prediction model is 0.504°Brix,which is 0.227°Brix lower than the multi attitude combination model.The root mean square error of external validation on the machine is further reduced to 0.683°Brix,which has stronger application potential.