Research On Multi-mode Detection Method And Device Of Fruit Near Infrared Spectroscopy

Near infrared spectroscopy detection technology has been maturely applied to fruit quality detection.The fruit detection device can quickly and harmlessly detect the internal quality of the fruit.At present,not only the price of fruit detection device is relatively high,but the detection method of the detection device is single.The detection methods for different fruits cannot be met,and different fruit detection devices need to be used at this time.At the same time,the single detection method reduces the practicability of the fruit detection device and increases the purchase cost of fruit wholesalers.Therefore,the fruit detection device urgently needs to realize multi-mode universalization.At present,there are two structures for diffuse reflection,the 45°light source diffuse reflection structure and the ring light source diffuse reflection structure.Therefore,it is urgent to study which structure is better.Currently,diffuse reflection detection has an optimal detection distance,and it is more sensitive to fruit size.Therefore,there is an urgent need to study diffuse reflection devices with adjustable detection distances.This paper has carried out research work on issues such as"the detection method of each fruit detection device is single","which is the best 45°light source diffuse reflection structure or ring light diffuse reflection structure"and"the optimal detection distance exists for diffuse reflection detection".First,develop two devices"fruit detection device that can provide multiple detection methods"and"diffuse reflection detection device with adjustable detection distance".Then,the feasibility of the two devices was verified through the Apple SSC detection experiment,and the performance of the two devices was compared in the diffuse reflection detection mode to predict the performance of the apple soluble solids model.Finally,the prediction model of Apple SSC under three detection methods is established.The optimal apple diffuse reflection detection model and the general model of different sizes of apples at the optimal detection distance.The quantitative relationship between Apple's SSC and spectrum is obtained,which provides basic data and models for the future multi-mode dynamic detection device and the diffuse reflection dynamic detection device with adjustable detection distance.The main research contents and conclusions are as follows:(1)Study the feasibility of multi-mode fruit detection device that can provide multiple detection methods to detect apple SSC.The results show that the three kinds of spectra are preprocessed by GFS smoothed by 3-point Gaussian filtering to establish a model for predicting the SSC content of apples and obtain good results.The Rpre and RMSEP of the diffuse transmission model are 0.967 and 0.507°Brix,respectively;the Rpre and RMSEP of the total transmission model are 0.957 and 0.574°Brix,respectively;the Rpre and RMSEP of the diffuse reflection model are 0.949,0.536°Brix,respectively;Furthermore,diffuse reflectance spectroscopy is combined with CARS algorithm to screen characteristic wavelengths,and the Rpre and RMSEP of the model using 119 variables are 0.977 and0.362°Brix,respectively.30 Bingtangxin apples that did not participate in the modeling were used to test the model to predict the performance of Apple SSC.The correlation coefficient of the final 30 external verification sets is 0.906,and the verification root mean square error is0.707°Brix.This study verified the feasibility of the multi-mode device to provide multiple detection methods,and also established a sugar content prediction model for apples under different detection methods.(2)Investigate two diffuse reflection optical structures,which is better,45°optical structure and ring optical structure.First,the absorbance repeatability of the two visible/near-infrared detection systems in the 550?900nm spectral range was calculated;then the diffuse reflectance spectra of 220 apples were collected and converted into absorbance spectra.Finally,the CARS and UVE algorithms were used to screen the effective variables of the original spectrum in the 550-900nm spectral range,and the effective variables and apple soluble solid content were used to establish a PLS model.The results showed that the absorbance repeatability of 45°optical structure and ring optical structure were 0.0017AU and 0.0008AU respectively in the 550?900nm spectral range,and the absorbance repeatability under ring optical structure conditions was better.The model established by the CARS algorithm to select variables is significantly better than the model established by the UVE algorithm.The CARS-PLS models established under the conditions of 45°optical structure and ring optical structure were compared.The Rp and RMSEP of 45°optical structure using57 variables to build the model were 0.843 and 0.724°Brix,respectively.The Rp and RMSEP of ring optical structure using 93 variables to build the model were 0.924 and 0.475°Brix,respectively.Obviously,the prediction accuracy of the ring optical structure model is higher,which is consistent with the results of absorbance repeatability under the two conditions.The CARS-PLS model established by ring optical structure was further verified externally,and the correlation coefficient of the verification set and the root mean square error of the verification set were 0.879 and 0.517°Brix,respectively.This study shows that the effective information collected by the 45°optical structure optical structure is concentrated in the visible light band,which will lead to poor robustness of the model.The optical structure of ring optical structure reasonably reduces stray light,so it can collect more effective information related to Apple SSC,and the established model is of higher quality.This study verifies that the annular diffuse reflection structure is better.(3)Research the optimal detection distance of the adjustable diffuse reflection device when detecting apples,and establish an apple size compensation model to improve the robustness of the Apple SSC prediction model.According to the size,apples are divided into small fruits(66-75mm),medium fruits(75-80mm)and large fruits(above 80 mm).Adjust the distance from the center of the fruit tray to the surface of the probe,so that the distance from the center of different sizes of apples to the surface of the probe is 4cm,50mm,and 60mm.Establish an SSC prediction model for three fruit types at three distances.And respectively select the spectral data of the three fruit types at the optimal distance.Under the optimal distance,the spectrum of the mixed three fruit types is pre-processed and band-selected,combined with the partial least square method to establish an apple SSC prediction model.The results show that the three fruit types have the best performance in establishing the model under the condition of 50mm,with Rp~2 and RMSEP values of 0.881 and 0.551°Brix,respectively.The mixed fruit spectra are preprocessed by baseline offset correction&Savitzky-Golay smoothing(BOC-SGS)and competitive adaptive reweighted sampling(CARS)band selection.94 valid variables are used to establish a Partial least squares regression(PLSR)model to predict the second batch of small The Rp2 and RMSEP of the fruit are 0.805 and 0.748°Brix,respectively.Compared with the original spectral prediction model,the Rp2 is increased by 1.8%,and the RMSEP is reduced by 3.4%.It is predicted that the Rp2 and RMSEP of the second batch of fruit are 0.861 and 0.581°Brix,respectively.Compared with the original spectrum prediction model,the Rp2 is increased by 4.7%,and the RMSEP is reduced by 9.2%.The Rp~2 and RMSEP of the second batch of large fruits are predicted to be 0.726 and 0.727°Brix,respectively.Compared with the original spectrum prediction model,the Rp~2 is increased by 2.7%,and the RMSEP is reduced by 3.6%.The hybrid model is not sensitive to changes in fruit size,which improves the robustness of the apple SSC prediction model.