Comprehensive Analysis of Fruit Detection, Ripeness Assessment, and Mass Estimation Using YOLOv7, ResNet, and ViT Models

China became the world's largest apple producer, accounting for over 50% of the global planting area and output. The increasing urbanization has led to a decline in agricultural workers and higher labor costs, prompting the inevitable trend of employing picking robots for apple harvesting, supported by the widespread use of computer vision technology in agricultural automation. This paper employs a comprehensive approach using YOLOv7, ResNet, and ViT models for fruit detection, ripeness assessment, and quality estimation. Initially, the YOLOv7 model processes input images to generate detection boxes identifying apples and further quantifies the number of apples in each image. Subsequently, based on the detection boxes from the YOLOv7 model, color analysis is performed to calculate fruit ripeness, presenting the results in the form of a histogram for a more intuitive understanding of ripeness distribution. Next, by calculating the red pixel area of identified apples, the study estimates fruit quality and presents the results as a histogram, illustrating the distribution of fruits of varying quality. Finally, the ResNet and ViT models are used for image classification, comparing their accuracy on a validation set to evaluate their performance differences in fruit image classification tasks. This integrated process provides a comprehensive and accurate analysis for fruit detection and assessment, offering robust support for subsequent research in fruit quality analysis and classification.