An effective approach for multi-class fruit and vegetable classification based on modified ResNet-18 from a combined dataset

Abstract

This study delves into the pivotal role of fruit and vegetable quality in influencing human health, emphasizing the necessity of prioritizing fresh produce for optimal well-being. Consumption of high-quality fruits and vegetables offers enhanced nutrition and antioxidant benefits, contributing to overall health. Conversely, consuming spoiled produce poses health risks due to exposure to harmful pathogens and potential loss of essential nutrients. Deep learning emerges as a promising approach for assessing fruit and vegetable quality, leveraging its capacity to discern nuanced characteristics from complex datasets. Neural networks, particularly deep learning models, excel in automatically identifying subtle visual cues indicative of freshness or decay. This technology facilitates rapid and accurate categorization, essential for ensuring the quality and safety of produce across various settings, from farms to retail outlets. In this research, we amalgamated and analyzed two datasets to create a comprehensive dataset. Subsequently, we devised a new labeling system and applied three image filters for thorough analysis. To classify images as either good or rotten quality, we proposed a modified ResNet18 model, fine-tuned through hyperparameter optimization. Our findings showcase the effectiveness of the proposed model, achieving a remarkable 98% accuracy with an image size of 224x224 and a batch size of 64 over ten epochs. This model holds potential for assisting health-conscious individuals in discerning the quality of fruits and vegetables, thereby aiding in informed dietary choices. Overall, this study underscores the significance of fruit and vegetable quality in promoting human health and underscores the utility of deep learning in ensuring produce safety and quality assessment.