Study On Detection Method Of Sesame Oil Adulteration Based On Near Infrared Spectroscopy And Pattern Analysis

In recent years,adulteration of edible oil is a food quality problem in China,which has received great attention from the whole society.The practice of incorporating low-value oils into high-value edible oils seriously undermines the legitimate rights and interests of consumers.At present,the adulteration detection of edible oil mainly adopts physical and chemical methods.These methods often take a long time and have complicated operation process,which is not suitable for large-scale adulteration detection of edible oil.Therefore,it is a significant task to explore a efficient method for adulteration detection of edible oil.Near-infrared spectroscopy has been applied in petroleum,agriculture,cottonocracy and other fields.Compared with traditional physical and chemical analysis methods,near-infrared spectroscopy has the advantages of high sensitivity,high stability and fast online analysis.Based on near-infrared spectroscopy,this thesis uses pattern analysis technology to make further exploration in the field of edible oil adulteration detection.Taking sesame oil adulteration as an example,Sesame oil was used as the base oil.Soybean oil and rapeseed oil were used as adulterated oil respectively.372 of each type of adulterated samples are configured according to 31 concentration gradients.Then near-infrared spectral data of samples were collected.In solving the problem of identification of adulterated species of edible oil,K-SVD dictionary learning algorithm combined with support vector machine was used to establish a adulteration identification model to identify the adulterated sesame oil and the category of adulterated oil.Then this thesis establish a PCA-SVM classification model and a single support vector machine classification model as comparison.The feasibility and effectiveness of K-SVD dictionary learning algorithm in spectral feature extraction is also explored.In order to solve the problem of detecting the content of adulterated oil in edible oil,the wavelength variable participating in the modeling is carefully selected after rough selection.After screening the wavelength variables by the uniformative variable elimination(UVE),this thesis attempts to make a more in-depth analysis of the wavelength variables suitable for establishing the adulteration quantitative detection model.The extremum disturbed and simple particle swarm optimization(tsPSO)and the Synergy interval partial least squares(SiPLS)were used to establish the adulteration quantitative detection model.The most suitable combination of characteristic bands for adulteration quantitative detection model of sesame oil was found.Finally,Lasso regression is used instead of partial least squares regression to establish the UVE-Lasso adulteration quantitative detection model.This method can not only avoid over-fitting,but also filter wavelength variables,simplify the model and improve predictive efficiency.The results showed that the established three edible oil adulterative identification models were able to accurately identify the adulterated sesame oil and the category of adulterated oil.Among them,the K-SVD dictionary learning algorithm combined with the support vector machine model has achieved 100% prediction accuracy on both the training set and the test set,which is better than the other two models.It can be seen that the K-SVD dictionary learning algorithm can extract more complete spectral information than PCA,and the learned dictionary can be used to establish the adulterative qualitative identification model to improve the efficiency and accuracy of the model.the uniformative variable elimination method is used to initially screen the wavelength variables,and then the tsPSO is used to optimize the selected feature band combination in SiPLS to establish an optimal adulteration quantitative detection model.The prediction error of the adulterated content is the smallest in this model.Finally,this thesis use UVE combined with Lasso regression to establish the adulteration quantitative detection model.Although the improvement of prediction accuracy is not obvious,but the characteristic wavelength variable used for modeling is greatly reduced.At the same time,the risk of overfitting is reduced.