Dynamic On-Line Detection Of Apple Defect And Sugar Content By Near Infrared Spectroscopy

For many years,Chinese fruits output ranks first in the world,but it is only the fourth fruit exporter,and has been in an awkward position of trade deficit.The reason is that the commercialization level of fruits in China is low,and the proportion of high-end fruits is relatively less.The mixed packaging and mixed selling of fruits make it difficult for high-quality fruits to achieve high prices.There is still a certain gap in the added value of fruits compared with the developed countries,which leads to the weak competitiveness of Chinese fruits industry in the international fruits market,faced tremendous market pressure.In order to get rid of the dilemma,it is necessary to improve the quality and commercialization level of domestic fruits and complete the transformation from "low price competition" to "quality competition".Relying on near infrared dynamic online detection technology,fruits can be classified according to quality quickly,nondestructive and accurately,and fruits can be priced according to quality.Overcoming the huge disadvantages of traditional manual sorting,such as the strong subjectivity,low efficiency,and less ability of recessive defects classification,which has significant practical significance for improve the quality of fruit in our country,commercialization levels as well as international market share.Throughout the research progress in the field of dynamic online detection at domestic and overseas,the detection objects and detection accuracy have been better improved,which greatly promoted the trend of near-infrared dynamic online detection technology to become the mainstream fruit quality detection technology in the future.However,the detection index is mainly sugar content,which is still less involved in defect detection.Moreover,the dynamic online detection of fruit defect and sugar content has not been achieved,which cannot meet the multi-facet requirements for fruit quality in the international fruit market.Therefore,aiming at the problem of fruit sorting,apples from Aksu,Xinjiang,and Red fuji from Luochuan,Shaanxi were selected as the research objects in this paper.Near infrared diffuse transmission spectra of apple samples were collected by the dynamic online detection device of fruit quality,and the NIR dynamic online detection of apple defects and sugar content were carried out.This paper involves three main research contents a nd conclusions are as follows:(1)Apples from Aksu,Xinjiang,with frequent occurrence of water core,were taken as the research object,carried out the dynamic online detection of apple internal defect water core and soluble solid content.After comparing and analyzing the spectral characteristics between water core apples and sound apples.The results showed that the accumulation of sorbitol solution in water core apples,resulted in the decrease of the air space between cells and the weak light scattering ability,which was the main reason for the differences of the spectrum between apples with or without water core.The qualitative models of PCA,PLSDA and LSSVM for internal defect water core apples and sound apples were established by using stoichiometry.After comparing the model accuracy,PLSDA model with linear correlation characteristic was selected,which could distinguish apples with water core from sound apples.Then,considering the influence of mixed apples with internal defect water core and sound apples on the accuracy of the model,and considering the rationality of the actual fruit sorting process,sugar content PLS model were established after separating apples with water core and sound apples.For the water core apples,the correlation coefficient and root mean square error of calibration set of the sugar content quantitative model were 0.96 and 0.21%,while the correlation coefficient and root mean square error of prediction set were 0.86 and 0.37%.For the sound apples,the correlation coefficient and the root mean square error of calibration set of the sugar content quantitative model were 0.94 and 0.35%,while the correlation coefficient and root mean square error of prediction set were 0.83 and 0.50%.After the qualitative identification model of internal defect water core and the quantitative prediction model of sugar content were loaded into the dynamic online sorting device of fruit quality,the results of external verification showed that the dynamic online sorting accuracy of internal water core defect and sugar content were 94.29% and 91.43% respectively.(2)Red Fuji Apples from Luochuan,Shaanxi,which is pellicular and easy damaged,were taken as the research object,carried out the dynamic online detection of apple external defects bruise and soluble solid content.After comparing and analyzing the spectral characteristics between bruised apples and sound apples.The results showed that surface hardness and the slight darkening of the surface color of the bruised apples,affected the absorption of water and color by the near-infrared light,which were the main reasons for the differences of the spectrum between bruised apples and sound apples.The qualitative models of PCA,PLSDA and LSSVM for external defect bruised apples and sound apples were established by using stoichiometry.After comparing the model accuracy,PLSDA model with linear correlation characteristic was selected,which could distinguish bruised apples from sound apples.Then,considering the influence of mixed external defect bruised apples and sound apples on the accuracy of the model,and considering the rationality of the actual fruit sorting process,sugar content PLS model were established after separating bruised apples and sound apples.The correlation coefficient of prediction set increases from 0.85 to 0.96,and the root mean square error of prediction decreases from 0.60 to 0.27.After the qualitative identification model of external bruise defect and the quantitative prediction model of sugar content were loaded into the dynamic online sorting device of fruit quality,the results of external verification showed that the dynamic online sorting accuracy of external defect bruise and sugar content were 100% and 92% respectively.(3)Four kinds of apples with different bruised time(12h,24 h,36h and 48h)were taken as the research objects,carried out the dynamic online detection of apples with different bruised time classification.After comparing and analyzing the spectral characteristics of apples with different bruised time.The results showed that the change of color and moisture with time were the main reasons for the differences of apple spectrum with different bruised time.The qualitative models of PCA,PLSDA and LSSVM for apples with different bruised time were established by using stoichiometry.And the three models were compared and analyzed,results show that: the PCA model was basically unable to distinguish the apple samples with four different bruised time,with the increase of bruised time,the overall misjudgment rate of PLSDA and LSSVM models showed a decreasing trend.The least square support vector machine model established by the RBF?Kernel with nonlinear characteristic,performed a better classification effect than other linear models in predicting different bruised time.The prediction accuracy of apples with 12 h,24h,36 h and 48 h bruised time was 86.67%,66.67%,100% and 100%,respectively.After the qualitative identification model of apples with different bruised time was loaded into the dynamic online sorting device of fruit quality,the results of external verification showed that the apples with bruised time of 36 h and 48 h can be accurately pushed into the corresponding classification exit,which achieves a relatively ideal effect of dynamic online classification of apples with different bruised time.