AUTOMATED QUALITY ASSESSMENT OF APPLES USING CONVOLUTIONAL NEURAL NETWORKS

Abstract

Quality assessment of apples is a pivotal task in the agriculture and food industries, with direct implications for economic gains and consumer satisfaction. Traditional methods, whether manual, mechanical or electromechanical, face challenges in terms of labor intensity, speed, and quality control. This paper introduces a solution using machine learning algorithms – specifically, Convolutional Neural Networks (CNNs) – for a more nuanced and efficient apple quality assessment. Our approach offers a balance between the high-speed capabilities of electromechanical sorting and the detailed recognition achievable with human evaluation. A dataset consisting of over 2000 apple images, labeled as 'Good' or 'Damaged', was compiled for training and validation purposes. The paper investigates various architectures and hyperparameter settings for several CNN models to optimize performance metrics, such as accuracy, precision, and recall. Preliminary evaluations indicate that the MobileNet and Inception models yield the highest levels of accuracy, emphasizing the potential of machine learning algorithms to significantly enhance apple quality assessment processes. Such improvements can lead to greater efficiency, reduced labor costs, and more rigorous quality control measures.