Implementación de un Sistema de Visión Artificial para la Detección de Plagas en las Plantaciones de Tomate

This work presents the development of a computer vision system based on the YOLOv8 model, oriented to the automatic identification of pests in tomato crops, focusing specifically on Prodiplosis longifila and the whitefly (Bemisia tabaci). For this purpose, 800 images were collected under real field conditions in the locality of Huanchaco, La Libertad, under a controlled environmental setting. The data set was divided into 70% for training, 20% for validation and 10% for testing. The YOLOv8n model was trained for 80 epochs, applying data augmentation techniques and the AdamW optimizer. The results achieved indicated an accuracy of 38.40% for Prodiplosis longifila, 12.8% for whitefly and 47.2% for pest-free tomatoes, with corresponding F1-scores of 44.1%, 14.90% and 47.3%. Although the system achieved a processing speed of 24 FPS, the limited accuracy reflects the challenges involved in detecting microscopic-sized pests in real environments. This study lays a methodological foundation that will allow for future improvements, identifying the main current limitations and aspects that need to be optimized to achieve adequate performance in commercial applications.