Studies On The Adulteration Modeling Analysis And Its Implementation For Grape Seed Oil

Grape seed oil is an important food and health care resource.For the sake of high profit,grape seed oil on the market is often mixed with other low-cost vegetable oil or even waste oil to obtain high profits.Adulterated grape seed oil will seriously affect the health of consumers and the health of edible oil industry.Therefore,it is of great significance to study a fast and highly sensitive dynamic grape seed oil analysis technology to supervise the quality of oil enterprises and to protect the rights and interests of consumers.Near infrared spectroscopy is a green,rapid,and nondestructive detection technique developed in recent years.By establishing a model of the relationship between the near-infrared absorption peaks of a substance and the content or the composition of the substance,the properties of the chemical composition can be accurately predicted.However,due to the complex composition of oil,the near-infrared spectral data of the sample in the whole band are with many wavelength points and serious data superimposition,thus it is difficult to establish a stable and reliable qualitative analysis model.In this paper,the data of near-infrared spectra of pure grape seed oil,grape oil blended with different proportions of other oils(peanut oil,corn oil,soybean oil and sunflower oil)were taken as research object.The pretreatment method,the wavelength optimization of the near-infrared spectra of grape seed oil,and the rapid identification model of grape seed oil adulteration were studied.On the basis of theoretical researches,an online analysis system grape seed oil adulteration identification was developed to achieve non-destructive and fast testing of the adulterated oil.The main research work is as follows:(1)Aimed at the problems of large amount of near-infrared spectral wavelength points and severely overlaid data,a firefly optimization algorithm combined with a characteristic wavelength optimization method of continuous projection algorithm is proposed to reduce the complexity of adulteration discriminant model.Firstly,five types of spectral data were preprocessed by using multivariate scatter correction preprocessing method.Then,the firefly optimization algorithm and the continuous projection optimization algorithm were used to select the near-infrared spectral wavelength for the rough selection stage and the second-order optimization stage,respectively.And the characteristic wavelength points most relevant to the classification results were obtained.Experimental results show that the proposed algorithm selects 17 most relevant wavelength points and greatly reduces the dimension of the modeling data.(2)In order to improve the identification accuracy of adulterated species,a novel identification method for the adulterated grape seed oil species was proposed.Firstly,based on the five groups of near-infrared spectroscopy.data,the extreme learning machine(ELM)classification model was established to obtained classification results.Then,considering that misclassification problems that may occur in the ELM classification model or new samples that do not belong to the modeling database category may be collected,the clustering algorithm with clustering center automatically determined was utilized to further judge the classification results of ELM model.By calculating the numbers of clustering centers in each classes of the classification results,the classification accuracy can be reconfirmed.(3)Based on the theoretical methods,an online detection system of grape seed adulteration was developed and was applied to the identification of grape seed oil adulteration types.The hardware components of on-line inspection system and the realization of analysis software are designed.The performances of the system are verified.The experimental results show that the on-line testing system can effectively detect the adulterated grape seed oil.