Rapid and Non-Destructive Predicting the Ash content of Wheat Flour using a Computer Vision System

In this study a computer vision system was employed to investigate the relationship between the image features and ash content of wheat flour. Image features of flour surface included color properties (L*, a*and b*) and gray level co-occurrence matrix (GLCM) parameters (contrast, energy, correlation, homogeneity and entropy). Results of correlation analysis revealed that there were significant linear relationships between image features of surface flour (except for correlation) and their ash content. However, because of low coefficient of determination (R2) of linear models, quadratic models were fitted to data in order to predict ash content of wheat flour. Analysis of variance showed that the fitted quadratic models, except for the correlation, were significant. As well, the R2 values of significant models, except for L*, a* and energy were satisfactory (R2>0.75). Results of models validation showed that proposed quadratic models had good performance to predict ash content of new wheat flour samples.