Quality Detection Of Yellow Peach Chips Based On Computer Vision And Visible/Near-Infrared Spectroscopy

Peach(Prunus persica(L.)Batsch)is one of the most favorable fruits in the world owning to its pleasant flavor and high nutrient content.However,peach can easily be damaged with mechanical bruise and pathogenic spoilage fungi,and not suitable for long time storage.Undoubtly,as meeting the demand of market and consumers for low fat,low calorie and high cellulose of peaches,peach chips has been processed via a kind of the dry-processed procedures.During the procedures,the moisture of the fresh fruit was changed sharply,while also with other quality attributes variation,like size,color,soluble solids content(SSC)and firmness.In order to evaluate the results and fruit chips quality affected by extraneous physiological and psychological factors.The analytical methods for peach chips detection are required that would monitor the quality changes.Traditionally,analytical methods for determining chips mainly rely on subjective experience judgments.Although those methods are precise,they are tedious,time-consuming,destructive,inefficient and hard to satisfy the need of fast,nondestructive detection in chips industry.Hence,it is of great importance to develop a rapid and nondestructive technology to detect the fruit chips,improve the international competitiveness of fruit chips industry in China and even can help to develop a wider international market.This research was supported by technology project of Jiangsu province's(project no.BN2015025)and graduate students' practical innovation project of Jiangsu province's(project no.2017).In this work,we chose yellow peach chips as the research object,which is one of most popular peach chips in the market.Computer vision technology was applied to detect external qualities,and in order to detect the yellow peach chips fast,a software was designed.Then,visible/near-infrared spectroscopy was used to detect internal qualities.So,this work can provide a theoretical basis to develop online detection equipment for peach chips.The main research contents were as follows:1.External quality detection of yellow peach chips based on computer visionThe image acquisition system for yellow peach chips was developed.The system was mainly composed of the industrial camera(include lens),light source,dark box and computer.Images of yellow peach chips were acquisited by the software of camera,then images were preprocessed by the software of Matlab.By the method of threshold segmentation,the dominate parameters including size,L*value,a*value and b*value were extracted from the interest region of image.The linear model of one element achieved good predicting performance,with Rp=0.92,RMSEP=1.57 cm2 for prediction area,Rp=0.87,RMSEP=2.27 for b*value,respectively.The SVM-C model was superior to the PLS-DA model for the classification of overall external qualities of yellow peach chips.The overall accuracy rate of SVM-C and PLS-DA models were 90.52%and 86.26%,respectively.A software was developed for external quality detection and classification of yellow peach chips,which develop the industrial camera software twice and was written by C#based on windows operating system.The sample images were acquisited by camera connected with USB,and the images were calculated and analyzed by computer.Based on the results of the experiment,the linear model of one element was applied to detection model and SVM-C model was applied to classification model.One hundred and five samples were used as external samples to exam the performances of device and established models.The results showed that the size,b*value could be well predicted with Rp=0.94,RMSEP=1.43 cm2 for prediction area,Rp=0.85,RMSEP=2.14 for b*value,respectively.The overall accuracy rate of SVM-C was 88.56%.2.Internal quality detection of yellow peach chips based on visible/near-infraredspectroscopy.The effects on PLS and LS-SVM of different spectral pretreatment methods such as standard normal variate transformation(SNV),multiplicative scatter correction(MSC),moving-average smoothing(MS)and 1 st-derivative(1-Der)for Vis/SWNIR and NIR were studied.The results showed that the best pretreatment method was MSC.For the prediction of SSC of yellow peach chips,the LS-SVM model based on Vis/SWNIR was superior to the LS-SVM model based on NIR for the prediction of SSC with Rp,RMSEP and RPD values of 0.761,1.998%and 1.532,respectively.For the prediction of firmness of yellow peach chips,the LS-SVM model based on NIR was superior to the PLS model based on Vis/SWNIR for the prediction of firmness with Rp,RMSEP and RPD values of 0.862,0.292 Kg and 1.991,respectively.To simplify the model,successive projections algorithm(SPA)was used to select characteristic wavelengths of SSC and firmness of yellow peach chips,and the new optimal LS-SVM models were developed for each parameters prediction.According to the prediction results of the full spectrum the wavelengths,seventeen characteristic wavelengths were selected from one thousand eight hundred and fifty-three full wavelengths of Vis/SWNIR based on SPA method of SSC,then SPA-LS-SVM model achieved the prediction for SSC with Rp,RMSEP and RPD values of 0.682,2.296%and 1.332,respectively.Moreover,fifteen characteristic wavelengths were selected from one thousand five hundred and fifty-seven full wavelengths of NIR based on SPA method of firmness,then SPA-LS-SVM model achieved the prediction for firmness with Rp,RMSEP and RPD values of 0.805,0.336 Kg and 1.752,respectively.