Estimating ash content in wheat flour using visible-near infrared hyperspectral imaging and machine learning methods

The quality of baked products is highly depending on the quality of wheat flour. Ash content is a key indicator of the flour extraction rate and plays a crucial role in evaluating flour quality and classification. Traditional methods for determining flour ash have destructive nature, time-consuming processes, and require skilled persons. In the present research, visible-near infrared hyperspectral imaging was employed as a non-destructive approach to estimate the ash content of wheat flour. The samples were prepared from the flour of Tufton and Berberd breads and their ash levels were determined through standard chemical tests. After acquiring hypercubes, they were processed by developing an algorithm in MATLAB software. The selected effective wavelengths using principal component analysis for Taftoon flour were 493.84, 652.59, 725.35, 826.23, 872.53, 894.85, and 944.46 nm, and for and Barbari flour were 422.73, 601.33, 746.85, 789.84, 803.07, 896.51, and 941.16 nm. Predictive models were then developed using artificial neural networks and partial least squares regression. The performance results of the two models showed prediction accuracies of 98.96 and 97.05% for Tafton flour, and 99.55 and 92.27% for Barbari flour, respectively. The findings also demonstrated that combining hyperspectral imaging with machine learning models can be applied for the accurate, rapid, and non-destructive estimation of flour ash content. This integrated approach presents a promising alternative to traditional quality control methods in food processing industries. • Hyperspectral imaging was applied to estimate the ash content of wheat flour. • The effective wavelengths and channels were selected using PCA method. • Efficient features were extracted from the hypercubes’ effective channels. • ANN model achieved higher prediction accuracy in Barbari and Tafton flour.