Non-Destructive Prediction of Bread Staling Using Artificial Intelligence Methods

Abstract

In foods with limited shelf life and in new product development studies, it is important for producers and consumers to estimate the degree of staling with easy methods. Staling of bread, which has an essential role in human nutrition, is an important physicochemical phenomenon that affects consumer preference. Costly technologies, such as rheological, thermal, and spectroscopic approaches, are used to determine the degree of staling. This research suggests that an artificial intelligence-based method is more practical and less expensive than these methods. Using machine learning and deep learning algorithms, it was attempted to predict how many days old breads are, which provides information on the freshness status and degree of staling, from photos of whole bread and pieces of bread. Among the machine learning algorithms, the highest accuracy rate for slices of bread was calculated as 62.84% with Random Forest, while the prediction accuracy was lower for all bread images. The training accuracy rate for both slice and whole bread was determined to be 99% when using the convolutional neural network (CNN) architecture. While the test results for whole breads were around 56.6%, those for sliced breads were 92.3%. The results of deep learning algorithms were superior to those of machine learning algorithms. The results indicate that crumb images reflect staling more accurately than whole bread images.