FruitVision Inteligencia Artificial para Clasificación

This article describes the development of a machine vision system called FruitVision, designed for the automatic classification of fruit according to its ripeness stage in agro-industrial contexts. The research arose from the problems associated with manual post-harvest inspection processes, characterized by subjectivity, inter-operator variability, and lengthy evaluation times, which lead to economic losses, commercial rejections, and limited traceability in the fruit supply chain.

The proposed system is based on deep learning techniques for real-time object detection. A complete data engineering workflow was implemented, including image collection in markets and agricultural fields, collaborative labeling using annotation tools, and standardization of the dataset in a format compatible with YOLO models. Subsequently, a YOLOv8-based model was trained and optimized through continuous validation and evaluation with accuracy metrics, achieving over 90% accuracy in identifying ripe, green, and damaged fruit.

The system also integrates a web backend for image processing, results storage, and digital record generation, enabling traceability of the analyzed batches. The architecture includes cloud services for image storage and a relational database with access control, as well as a mobile-accessible interface for operators and supervisors. This implementation facilitates real-time monitoring and operational decision-making in quality control.

The results demonstrate a significant reduction in human error and batch inspection time, moving from lengthy manual processes to automated evaluations completed in minutes, while simultaneously improving the uniformity of the marketed product. Overall, the study demonstrates that the application of artificial intelligence in post-harvest classification is a viable alternative for increasing the efficiency, traceability, and competitiveness of the agribusiness sector.