Implementación de un sistema de clasificación de fresas empleando visión artificial para el control de calidad en función de sus etapas de maduración

Abstract

This study presents a strawberry classification system based on computer vision technology, focused on quality control according to ripeness stages. The proposed solution includes an image acquisition module consisting of an OV5647-62 camera and a Grove Vision AI V2 device, mounted on a mechanical structure designed to analyze strawberries as they move along a conveyor belt. To build the dataset, photographs were captured under controlled lighting conditions. The images were labeled and processed using the Edge Impulse platform to train a convolutional neural network. The resulting model was deployed on the embedded device through SenseCraft, enabling real-time system evaluation. The model’s performance was assessed using quantitative metrics such as precision, recall, and F1-score, achieving results above 97% in controlled environments. Additionally, tests were conducted under various lighting conditions, including controlled lighting, ambient light, and the absence of artificial light, to evaluate the system’s robustness to environmental changes. The findings indicate that the system is highly effective in classifying both ripe and unripe strawberries under appropriate lighting conditions, achieving 100% effectiveness during real-time testing. However, limitations were observed in scenarios without controlled lighting, highlighting the importance of environmental conditions for optimal system performance. This project demonstrates the feasibility of implementing an affordable computer vision system for agricultural quality control, enabling the automation of visual inspection processes and laying the groundwork for future optimization in real-world production environments.