Research On Comprehensive Evaluation Of Potato Quality Based On Information Fusion Technology

Potatoes have characteristics of foods and vegetables.Potatoes are known as the world's fifth largest commercial crop due to its rich nutrition.Some kinds of defects,such as green skin,sprouting,mechanical damage,and rot etc.would result in the quality decrease of potatoes during the period of harvest,storage and transportation.They affect the edible quality of potatoes.Therefore,it is necessary to detect the defects of potatoes rapidly and non-destructively so that to remove defective samples in time and guarantee the food safety.Meanwhile,market competitiveness of potatoes can be improved.In this study,nondestructive methods for quality evaluation of potatoes were investigated based on the data fusion of image information from machine vision and odor information from electronic nose(E-nose).The main contents of this research are briefly described as follows :1.Measurement of physical and chemical indexes of potatoesIn this study,Aureobasidium pullulans were identified as the main pathogenic bacteria using microbiological and molecular biology methods.This variety of bacteria was used as a pathogen for preparation of potato rot samples.Gas chromatography-mass spectrometer(GC-MS)was used to got gas distribution characteristics of various samples then to guide the construction of sensor array inside E-nose.The content of solanine in normal,green and sprouted samples were measured using high performance liquid chromatography,which would be used as data of control group for establishment of identification model with the machine vision and E-nose method later.2.Development of quality evaluation method for potatoes based on machine vision(1)A suitable machine vision device for image acquisition of potatoes was established.According to characteristics of potatoes,the upper part of the image capture light box was designed in a hemispherical shape to provide good diffuse reflection effect on the surface of potato sample,while the lower part was designed ina cylindrical shape such that uniform illumination effect is formed around sample.(2)Establishment of quality evaluation method for potatoes based on image information.Images of potato samples were captured.Extract color features and build recognition model.The quality of potato was determined qualitatively and quantitatively by color characteristic.The K-nearest neighbor(KNN)model for discriminating the potatoes early infected by the A.pullulans were established.The recognition rate was 93.75% in training set and 92.5% in prediction set.The BP-neural network(BPNN)model for discriminating the potatoes with green skin and early sprouts were established.The recognition rate were 97.44% in training set and96.10% in prediction set.The Support vector machine(SVM)model for the discrimination of defected potatoes were established.The recognition rate were 94%in training set and 92% in prediction set.The results show that the potato quality discriminant method based on machine vision technology has realized the detection of potatoes with early defects such as greening,sprouting and rotting.SVM model was also established for quantitative prediction of potato solanine content.The modeling results showed that the correlation coefficient(Rp)was 0.75179 and the root means square error of prediction set(RMSEP)was 138.6021.The rapid prediction of solanine content in potato was realized.3.Establishment of quality evaluation method for potatoes based on electronic nose(1)GC-MS was used as to measure the gas components of potato.On the basis of the results of GC-MS analysis and sensor characteristics,the sensor arrays were optimized by load factor analysis,eight sensors were selected to establish the electronic nose sensor array for potato quality detection.(2)Establishment of quality evaluation method for potatoes based on odor information.The stable values of the sensor arrays were used as the characteristic variables to build model for identifying potato quality based on odor information.The KNN for discriminating the potatoes early infected by the A.pullulans showed that the recognition rate of the training set and the prediction set was 90% and 85%.The BPNN for discriminating the potatoes with green skin and early sprout showedthat the recognition rate of the training set and the prediction set was 98.7% and94.87%.The SVM model for discriminating the potatoes with defects showed that the recognition rate of the training set and the prediction set was 96% and 90%.The results showed that the potato quality discriminant method based on E-nose technology has realized the detection of the potatoes early defects.The SVM model was also established for quantitative prediction of potato solanine content.The modeling results show that the Rp and RMSEP was 0.74867 and 156.6253.The method for identifying potato quality based on E-nose technology has realized the prediction of solanine content of potatoes.4.A novel method for quality evaluation of potatoes was proposed based on fusion technique combing machine vision and electronic nose.The feature variables obtained by the machine vision and electronic nose were fused at the feature layer.The integrated feature variables were used for the establishment of potato discriminant models.The quality of potato was determined qualitatively and quantitatively according to the change of image and odor characteristics.The model for discriminating the potatoes early infected by the A.pullulans showed that the recognition rate of the training set and the prediction set was 93.75% and 92.5%.The model for discriminating the potatoes with green skin and early sprout showed that the recognition rate of the training set and the prediction set was 100% and 97.44%.The model for discriminating the potatoes with defects showed that the recognition rate of the training set and the prediction set was 96% and 92%.All categories of defect samples were accurately classified.The SVM model was also established for quantitative prediction of potato solanine content.The modeling results showed that the Rp and RMSEP was 0.84153 and 162.3692.The results of fusion technology were compared with those obtained by single test.The results show that the potato quality detection method based on fusion technology,whether in the discrimination of quality or on the quantitative prediction of solanine were superior to single detection technology.In this study,machine vision and electronic nose were employed to judge the potato quality comprehensively and achieved good results by having physical andchemical indexes of potatoes measured as control group.Results showed that new methods developed in this study were feasible for evaluation of potato quality and data fusion technology can improve detection accuracy.The results of this study can provide the theoretical basis and test support for the industrial application of rapid and nondestructive detection method for potatoes quality.