Quality Detection Of Shelf Life Of Fresh In-husk Walnut Based On Spectrum And Image Information

With the progress of science,people's ideology changes,and people pay more and more attention to health.Known as the "Viva" and "Longevity Fruit" walnut,it is well received by people.In recent years,sales of fresh walnuts have increased year by year due to their higher nutritional value and unique taste.This article takes "gift No.2" green peel walnut as the research object,based on visible / near infrared spectroscopy,and computer vision technology,establish a shelf life quality detection model of green walnut,to effectively detect the moisture,color texture,and color of the outer green peel and the inner walnut kernel of green walnut Provide a theoretical basis for the establishment of an online.The main research contents and main conclusions are as follows:(1)Prediction and analysis of moisture content and and color of walnut green peel based on visible /near infrared spectroscopy.For a variety of pretreatments of spectral information,modeling and comparison found that for green skin moisture content moving average(Moving Average)is the best pretreatment method Rc~2 is 0.9365,RMSEC is 0.0118,Rp~2 is 0.9323,RMSEP is 0.0116.For the color change baseline correction(Baseline),the modeling effect is the best.For the four color parameters L *,a *,b *,and ?E,their Rp~2 is 0.8382,0.7887,0.7854,and 0.8182,respectively,and their RMSEP is 1.7246,0.8308,1.7888,and 1.4458,respectively.After extracting the characteristic wavelength after detecting the moisture content of green peel,the model found that the model based on the Competition Adaptive Reweighted Sampling(CARS)is better than the model based on the regression coefficient(Regression Coefficien,RC),and the two merged RC + Partial Least-Square(PLS)model built by CARS is better,with Rc~2 of 0.9204 and RMSEC of 0.0132.For the change of green skin color,after the characteristic wavelength is extracted by RC,the modeling accuracy is better than that of the full band(Full Spectrum,FS)has been reduced but there is not much change.The modeling effect of L * and ?E among the four parameters is relatively good.The FS modeling accuracy Rp~2 is 0.8382 and 0.8182,and the RMSEP is1.7246 and 1.4458,respectively.(2)Predictive analysis of walnut kernel moisture content and color based on visible / near infrared spectroscopy.Different pretreatments were carried out on the spectral information.The modeling comparison found that the first derivative(1-Der)is the best pretreatment method for walnut kernel moisture content,and the modeling accuracy is better,Rc~2 is 0.8178,RMSEC is 0.0187,Rp~2 It is 0.8002 and RMSEP is 0.0189.For the color change L * and b *,the original spectrum modeling effect is the best,Rp~2 is 0.7149 and 0.4969,and RMSEP is 2.9813 and 2.7516,respectively.For a * and ?E are the optimal modeling effects after Baseline processing.Rp~2 is 0.7204 and 0.6849,and RMSEP is 0.4749 and 2.3841,respectively.After detecting the moisture content of walnut kernels and extracting the characteristic wavelength,the modeling found that the model based on the CARS algorithm is superior to the model based on the random leapfrog(RF)algorithm.The CARS-based modeling result is the best,and its prediction model Rp~2 is 0.7800,RMSEP It is 0.0205.For the color change of walnut kernels,after extracting the characteristic wavelength through RC,the modeling found that FS modeling is the best.Based on the modeling results of FS for the four color indicators L *,a *,b *,and ?E,Rp~2 is respectively0.7149,0.7204,0.4969,0.6849,RMSEP were 2.9813,0.4749,2.7516,2.3841.(3)Judging the shelf life of green walnuts based on image information.Using the five feature parameters of texture feature energy,entropy,moment of inertia,correlation,and inverse gap in four directions(0 °,45 °,90 °,135 °),a total of 20 parameters are used to determine the shelf life of the external green skin After optimizing the feature parameters and establishing the least squares support vector machine(LS-SVM),it is found that the discriminant model based on CARS-LS-SVM is the best,and the discriminant accuracy rate for 1,5,and 9 days is 100%.The recognition rate is lower for days 13 and 17.Color features RGB and HIS models were used to extract the color features to discriminate the shelf life of walnut kernels.The LS-SVM shelf life discriminant model was established.The results show that the LS-SVM discriminant model based on color features can better determine the shelf life.The accuracy of judgment on days 1,5,and 9 is 100%,while the discriminant results of 13 and 17 days were relatively poor.(4)Correlation analysis of the moisture content of walnut green skin and walnut kernel moisture content,green skin color and walnut kernel color,the results show that the two are significantly related at 0.01 level.The modeling found that the modeling effect of walnut kernel moisture content is better,the prediction set correlation coefficient is 0.9515,and the RMSEP is 0.0129.