| Jinhua ham with long history,special flavor,bright in color and excellent quality,abound in Jinhua of Zhejiang province,is the most famous local product in Jinhua of Zhejiang province.It is famous at home and abroad for its color,fragrance,shape and taste.Jinhua ham's annual consumption of 4 million,exports more than 3,000 tons,total sales of ham more than 2 billion yuan,accounting for about 70% of the national market according to the statistics.However,there are differences in the grade of ham products due to the influence of raw materials,processing season,weather conditions,processing technology and other factors.Limited to the current technology,the grading operation of ham is mainly accomplished by professional technicians through sampling and smelling,while sensory evaluation is susceptible to subjective factors and fatigue.Electronic nose,as a kind of instrument imitating human smell,can be used to distinguish the comprehensive smell.Therefore,in this study,a special electronic nose for selfmade ham was used and relevant algorithms were combined to identify the rank of ham.Therefore,the main volatile odor components of different grades of ham were studied by means of gas chromatography-ion migration spectrometer,Then the gas sensor was selected based on the volatile components results,and a portable electronic nose platform was built for the detection of jinhua ham grade.The electronic nose was first used to detect the different grades of chopped ham and explore how to extract the key difference features;Then,based on different grades of ham processed by "three-sign method",how to improve the stability of electronic nose detection of jinhua ham;Finally,the different grades of ham in the production workshop were tested.How to improve the anti-interference ability of electronic nose.The main research contents,methods and conclusions are as follows:(1)Volatile components of jinhua ham were detected by gas chromatography-ion migration spectrometer,and the results were analyzed by gc-ims Library Search V2.2.1.The volatile compounds of jinhua ham mainly include aldehydes,ketones,esters,alcohols,olefins and pyrazines.Based on the results,relevant gas sensors can be selected.Moreover,the analysis results showed that the concentrations of volatile odor components of different grades of ham were different,which also provided theoretical support for the electronic nose to distinguish the grades of jinhua ham.The gas sensor was selected to build a portable electronic nose hardware platform for detecting different grades of ham based on the analysis results.(2)A portable electronic nose was used to detect volatile smells inside different grades of chopped jinhua ham,A feature selection method based on the filter-wrapper framework is proposed to detect ham signals with large amount of irrelevant and redundant information in electronic nose,It is applied to the processing of ham data of electronic nose detection to extract the key difference information of electronic nose feature set.Compared with other feature selection algorithms,the method proposed in this study not only improves the prediction accuracy of electronic nose,but also reduces the computational amount of electronic nose data modeling.Finally,the combination of SVM and the feature subset corresponding to the proposed algorithm has the best classification effect(prediction accuracy of test set is 96.06%),and the computation is reduced(17.32s).The electronic nose hardware platform built in this study can be used to detect different grades of ham,and the proposed feature selection algorithm has a good classification effect.(3)Portable electronic nose was used to detect volatile smells of different grades of ham attached to bamboo skewers to reduce damage to experimental samples.n this process,the ham smell attached to the bamboo stick will lose part of the information,resulting in poor data stability.Therefore,aiming at the poor stability of electronic nose data encountered in the detection process,a pattern recognition algorithm based on the double-layer integrated convolutional neural network is proposed to classify the ham grade.Compared with classification algorithms commonly used in electronic noses,such as support vector machines,logistian regression,knearest neighbor,The integrated convolutional neural network algorithm proposed in this study has good model fitting effect(prediction accuracy of training set is 98.33%),and has the best classification ability(prediction accuracy of test set is 87.50%).The electronic nose can effectively improve the stability of detecting volatile smells of different grades of ham attached to the bamboo stick,thus improving the predictive power of the electronic nose.(4)The portable electronic nose was used to test different grades of ham of the same batch in the laboratory and jinhua ham production workshop,to study the influence of environmental interference in the workshop and how to effectively reduce the influence caused by interference.The electronic nose data set after feature selection is used to train a variety of classification algorithms to establish predictive modeling and analyze based on the predicted results,the prediction accuracy of the model established by using the data set collected in the ham production workshop is lower than that of the model established by using the data set collected in the laboratory,indicating that the interference of the workshop environment has a great impact on the electronic nose.Therefore,this study proposed a data processing method based on discrete analysis to process the data set collected in the environment of ham production workshop.Then a variety of classification algorithms such as support vector machines,logistian regression,knearest neighbor and the integrated convolutional neural network proposed in this study are trained to establish the prediction model.It was found that the prediction accuracy of different classification algorithms was improved(SVM: 3.33%,LR: 2.50%,KNN: 8.34%,integrated convolutional neural network :3.34%).It is shown that the data processing method based on discrete analysis can reduce the interference in the ham workshop environment and improve the prediction accuracy of electronic nose. |
PDF Full Text Request