| Apple is an important agricultural product and trade commodity in China,with considerable market share.The detection and grading in terms of apple quality can increase its commercial values and improve the commercial processing level of apple market.With increasing apple production of China,a self-owned visible/near infrared spectroscopy quality detection system for apple can avoid importing expensive equipments from abroad and upgrade automation equipments in the apple industry.Furthermore,it will bring huge economic benefits and enhance the international competitiveness of the Chinese apple industry.Visible/near infrared(Vis/NIR)spectroscopy detection technology is widely used in apple quality detection because of its advantages of nondestructive,fast,low cost,simple operation,no sample pretreatment,green and environmental protection.It provides technical support for online detection of apple quality.In this study,Fuji apple harvested in China was used to develop an on-line soluble solids content(SSC)Vis/NIR detection system of apples based on bicone roller transportation.The construction and design of this system were introduced and the influences of apple detection posture and detection point number on the SSC prediction models of apple were studied.The main research contents and conclusions are as follows:(1)The composition of the on-line SSC Vis/NIR detection system used in this study was designed and constructed.The system mainly included mechanical transportation and sorting module,information processing and controlling module and Vis/NIR detection module.With the aim of designing robust Vis/NIR detection module,the diffuse reflection detection mode and the light source arrangement scheme were determined.150 W halogen light source and 1000μm optical fiber are selected,and the integral time was calculated to be 100 ms.The system preliminarily achieved on-line acquisition of Vis/NIR spectra,which laid the foundation for the establishment of the following SSC prediction models.(2)The influence of apple detection posture on on-line SSC prediction models was studied.Firstly,the characteristics of apple detection postures based on bicone roller transportation were analyzed and then divided into seven types:posture 1 to posture 7,of which the first six ones were fixed postures and posture 7 was random posture.Then the influence of each apple detection posture on SSC models was discussed with the help of spectral data analysis and internal and external cross-validation models.It was found that the model established by spectra acquired at posture 3 had the best accuracy when the light source and detector corresponded to the equatorial part of the apple at the same time.The model established by spectra acquired at posture 6 had the worst performance.The model established by spectra acquired at posture 7 had better accuracy when it was used as the calibration set.Besides,models established by the spectra acquired at symmetrical postures showed similar performance.At last,the average model and global model were established by using the average spectra of each posture and all spectra respectively,and the compensation effect of these models on posture influence was verified.The average model had better accuracy(Root mean square error of calibration,RMSEC:0.356oBrix;Relative coefficient of calibration,r_c:0.947;Root mean square error of prediction,RMSEP:0.370oBrix;Relative coefficient of prediction,r_p:0.906),while the global model had better stability and prediction ability(RMSEC:0.488oBrix;r_c:0.893;RMSEP:0.506oBrix;r_p:0.851),which can be expected to be applied in the practical production.(3)The influence of detection point number on on-line SSC prediction models was studied.Firstly an on-line SSC Vis/NIR double-point detection system was introduced based on binary-branch optical fiber.Then posture 3 and 7 were selected to acquire spectra using both single-point and double-point detection system.Partial least squares(PLS)and stepwise multiple linear regression(SMLR)models were established to illustrate the influences of detection point number on SSC prediction models.Furthermore,standard normal variate transformation(SNV)and multiplicative scatter correction(MSC)were utilized to remove useless information from Vis/NIR spectra and consequently improved the PLS and SMLR models.The optimal PLS model(RMSEC:0.497oBrix;r_c:0.8966;RMSEP:0.524oBrix;r_p:0.8085)was established by spectra acquired at random posture 7 using double-point detection system and utilized SNV as spectral preprocessing method.The optimal SMLR model(RMSEC:0.446oBrix;r_c:0.9166;RMSEP:0.458oBrix;r_p:0.8662)was established by spectra acquired at random posture 7 using double-point detection system and utilized MSC as spectral preprocessing method.The latter model was superior to the former one in terms of both accuracy and stability.The experimental results validated the feasibility of the double-point detection system’s compensation effect for the influence of apple detection posture on on-line SSC prediction models.It also provided a reference for improving the accuracy and stability of apple SSC Vis/NIR on-line detection system. |
PDF Full Text Request