On Detection Method of Wheat Quality Based on Image Processing and Support Vector Machine

Abstract

This paper studies the detection and grading method of wheat quality based on image processing and Support Vector Machine. By wavelet packet and morphological open-close operation, the author extracts the morphological, color and texture features from whole-grain images of wheat. The results show that the color feature can largely reflect the difference between different grades of wheat; morphological feature can also reflect the difference but not as sharply as the color feature does; texture feature has little to do with wheat grade. Hence, the color feature is taken as the basis for whole-grain grading of wheat. Then, the author establishes a linear parameter classification and recognition model and BP neural network classification model for wheat quality. The overall recognition rate of the former model reaches 95.5%, and that of the latter model reaches 97%. The characteristic parameters of wheat are extracted by CEEMD signal processing method, and subjected to regression analysis by the multiple linear regression model. The correlation coefficient of the regression model is put at r2=0.9511, indicating that it is feasible to detect wheat quality by the collision method.