Research On Feature Processing Methods In Nondestructive Testing Of The Species Of Cereals And Oils

Cereals and oil crops are the foundation of the country.The nutritional health and life safety of human beings are closely related to the quality and safety of grain and oil.The traditional detection methods of cereals and oils rely on manual sorting,qualitative analysis of traditional imaging and quantitative analysis of chemical determination.These methods have the disadvantages of time-consuming,high cost,easy detection of interference,and inaccurate detection results.And the requirements for the testing environment and testing personnel are very high.The chemical reagents used are poisonous and harmful,which will not only destroy the test samples,but also pollute the environment.Therefore,it is of great significance to find a kind of real-time online,efficient,accurate,green and non-destructive detection technology for grain and oil.This article is mainly to study the feature processing methods in the nondestructive testing of cereals and oils.For the identification of edible oils,it is mainly based on Raman and near-infrared spectroscopy techniques and the use of pattern recognition methods to identify the types of edible oils.Eight kinds of pretreatment methods were used to pre-process the Raman near-infrared spectroscopy data of five edible oils such as peanut oil,corn oil,rapeseed oil,sunflower oil and soybean oil.First,characteristic correlation methods such as canonical correlation analysis,CARS,SPA and other characteristic processing methods are used to obtain the characteristic wavelengths of Raman NIR spectral data after pretreatment.Then,using support vector machine classification and K-nearest neighbor method to establish the characteristic wavelength of the edible oil Raman and near infrared spectroscopy fusion identification model.The established species identification model can achieve 100% accuracy of species identification in the edible oil samples obtained from the current food collection,and can accurately identify the five edible oil types.The effect of the model based on canonical correlation analysis is better than the other two methods,because the model built by canonical correlation analysis has the fewest input characteristic wavelength variables under the same type recognition effect.Improve the efficiency and accuracy of the model,save the storage space of data,and reduce the cost of calculation.The goal of fast and non-destructive accurate detection of grain and oil types is achieved.For the problem of grain variety identification,the main solution is hyperspectral imaging technology combined with spectral characteristic wavelength selection.The sparse principal component analysis algorithm is used to select the characteristic wavelengths of the four hyperspectral images of the collected rice seeds,and then the K-means clustering is used to establish a rapid detection model of cereal varieties.Principal component analysis was used to extract the feature information of the hyperspectral image of cereal seeds,which was used to compare and verify whether the characteristic wavelength selected by the sparse principal component analysis was interpretable.The sparse principal component analysis(SPCA)full-band K-means clustering model has the best effect and the highest recognition rate is 90%.At this time,30 characteristic wavelengths are selected.The highest recognition rate of the model that directly establishes K-means clustering for the entire band is only 40%.When the hyperspectral spectral characteristic bands are grouped,the K-means clustering model established by the sparse principal component analysis has the best recognition rate of 81.25%.At this time,18 characteristic wavelengths are selected.The highest recognition rates of the established K-means clustering model of principal component analysis are 55% and 67.5%,respectively.It can be seen from the experimental results that the characteristic wavelength of the hyperspectral image of cereals extracted by SPCA is more explanatory,and it can identify the varieties of cereal seeds more efficiently and quickly,and improve the recognition rate of cereal varieties.