Assessing Fruit Freshness with Deep Learning: A Case Study on Banana Using GoogLeNet and Transfer Learning

Abstract

Freshness is a critical determinant of fruit quality, significantly impacting consumers' health and purchase decisions, as well as market pricing. Consequently, the development of effective methods for evaluating fruit freshness is of utmost importance. In this research, we focus on bananas as a case study and employ transfer learning to analyze the dynamics of freshness over time, establishing a relationship between freshness and storage duration. We leverage the GoogLeNet model to automatically extract relevant features from banana images, subsequently employing a classifier module for classification. Our results demonstrate that this model can effectively detect banana freshness at a level surpassing human assessment. To assess the model's robustness, we extend its application to monitoring the freshness evolution of apples, revealing its continued utility. Based on these findings, transfer learning emerges as a precise, non-destructive, and automated method for monitoring fruit freshness. This technique holds promise for broader applications in the realm of vegetable detection.