LAUSR

Non-destructive detection methods to identify food foreign contaminants meet the development of the food industry. Near infrared (NIR) spectroscopy and computer vision (CV) have been considerably intriguing due to the advantages of safety and rapidity. In this study, foreign contaminants, including metallic iron, polypropylene plastic, and hair in bread, were identified based on these two techniques. In NIR spectroscopy, the effectiveness of distance match and discriminant analysis methods combined with different spectral pretreatments was compared. It showed that the accuracy was 98%, 94%, 91%, and F-score was 0.97, 0.93, and 0.91 in the validation set for detecting foreign contaminants aforementioned when discriminant analysis combined with Savitzky-Golay smoothing was used. In CV, deep learning based on the modified U-net has applied to segment these contaminants on the surface of the bread; the accuracy of the test set was 95%, 93%, and 92%, respectively. Based on the results, it can be concluded that NIR spectroscopy and CV are both an operational way to detect foreign contaminants in bread. And these two techniques could be combined to apply in on-line detection further.