Automating Fruit Quality Inspection Through Transfer Learning and GANs

Abstract

The fresh fruit production industry is a multibillion-dollar market that faces numerous challenges, particularly during the later stages of production, such as quality control. Ensuring the quality of fresh fruit is crucial, as even a single unnoticed rotten fruit can compromise an entire batch. This paper aims to identify the deep learning model that achieves the highest accuracy in detecting rotten or subpar fruits. This paper evaluated a total of 12 models, including 10 transfer learning models and 2 custom convolutional neural networks (CNNs), along with a deep convolutional generative adversarial network (DCGAN) to augment data in classes without sufficient data, thereby enhancing model performance. Notably, some of these models achieved significant results, with accuracy exceeding 98%.