Research On Application Of Neural Network In Edible Oil Quality NIRSA

Edible oil is an important energy and nutritional source for our bodies, it is also an important part of the human diet. Its quality situation will affect the development of food industry, health security of consumers and social harmony and stability. At present, our country adopts mainly the traditional chemical methods to determine edible oil quality. It was affected by the personal factors, can not satisfy the requirement of the modern society for detecting edible oil quality which should be simple, rapid, accurate and on-site. In recent years, near infrared spectrum analysis technology is used widely in detection and analysis areas, its advantage is rapid, low consumption, no pollution, without sample pretreatment, and can meet multiple analysis requirement synchronously, more suitable for quality inspection and quality control in processing.Take the main quality parameters named soybean oil acid value and peroxide value as study objects, collected52acid value samples and42peroxide value samples, removed singular points according to the deviation between the actual values and predicted values, chose the absorption peak according to spectrum band regular pattern to establish models separated in order to select the best modeling wave band. To improve the precision of the model, compared the traditional pretreatment methods with the wavelet transform, found that Daubechies wavelets has a better denoising effect in preprocessing oil acid value and peroxide value spectrum.Focused on the application of artificial neural network in oil near-infrared spectrum analysis. It compared BP artificial neural network and RBF neural network with the traditional PLS modeling method, discussed how to choose the main parameters affecting network performance including the number of hidden neurons, momentum factor, learning rate, learning times and so on in the BP neural network. In RBF neural network the main parameters affecting network performance are also researched including hidden neurons, radial basis function density and display interval. The result demonstrated that artificial neural network is superior to the traditional PLS. The decision coefficient R2of oil acid and peroxide value achieved0.99127and0.99265.The relative standard deviation RMSEP of oil acid and peroxide value achieved4.01%and3.68%, the expected goal is realized. The feasibility of artificial neural network applied in oil NIRSA and the better modeling effect are proved.