A Computer Vision Approach for Classifying the Degree of Freshness in Fruits and Vegetables to Reduce Food Waste

The level of freshness of fruits and vegetables must be constantly monitored to reduce food waste and maximize its nutritional value. Existing approaches to this problem either consider a binary classification problem, where fruits and veg-etables are classified into the two classes of “fresh” or “rotten”, or a three-class problem, with “fresh”, “medium”, and “rotten” classes. This work challenges the three-class issue achieving an initial accuracy of 82%, using the VGG16 deep learning model, and 84%, with YOLO v5. After going through many experiments, we reached 94.58% test accuracy with VGG16, improving the current state-of-the-art performance by 11 percentage points. A detailed description of the experiments and the used algorithms, together with their hyper-parameters, is given in this paper to facilitate code reproduction.