Study On Optimization Of Grain Strength Classification Model Based On Multi - Layer SVM

Wheat occupies a very important position in food crops, it is widely used in many foods. Because wheat quality directly affects the quality of the food produced, it is more and more noticed by people. In recent years, as people focus on food safety, fast and efficient detection of wheat quality has developed into a hot research topic. Currently, researchers have made some achievements on this topic, such as the use of near-infrared spectroscopy technology to achieve the detection of wheat quality types. However, the current detection methods still have some problems, such as the model is not enough perfect, the accuracy of the detection model is not high. For studying the existing problems, this paper classifies wheat based on wheat gluten strength by using near-infrared spectroscopy. Meanwhile, this dissertation also in-depth studies of wheat types detection model, proposes multi-layer model to achieve the detection of wheat types classified by gluten strength.First, in order to make the selected samples is representative and ensure the accuracy of the original spectral data of experimental samples, this paper uses the method of mahalanobis distance to screen the collected spectral data twice. Meanwhile, the principal component analysis (PCA) is used to analyze spectral data and the feasibility of the gluten strength classification of wheat based on near-infrared spectroscopy has been demonstrated.Secondly, this paper studies the pretreatment methods of spectral data for handling the subsequent modeling data. This paper analyzes eight different methods which combined by the moving window average, first derivative, second derivative and standard normal variable transformation (SNV). Not only that, this article also analyzes three compression dimension methods, including PCA, partial least squares (PLS) and successive projection algorithm (SPA). Comparison shows that moving window average -the second derivative-SNV correction method and PLS compression dimension method is the best processing method for the spectral data of wheat.Finally, this paper studies the multi-layer classification model by using support vector machine (SVM). Based on the analysis of validation results of the wheat type (strong gluten, medium gluten, weak gluten and ordinary wheat) detection model established by the SVM, this paper puts forward the thought of multi-layer model to implement type detection. Two layer classification model based on the SVM achieves the type detection of wheat based on wheat gluten strength, the experimental results shows that the two layers classification model has good detection effect on four types of wheat samples, which achieved the type detection of wheat quickly and accurately.