Desarrollo de un sistema de visión artificial con redes neuronales convolucionales (YOLOv8n) para el grado de madurez del Ají Escabeche

In the context of agriculture, accurate ripeness classification of agricultural products such as pickled chili peppers is essential to optimize quality and reduce losses in the production chain. This study presents the development of a computer vision system using the YOLOv8 convolutional neural network to automatically identify and classify pickled chili peppers into three phenological stages: immature, pintón, and ripe. A total of 3,021 manually labeled images were collected under real growing conditions, without applying data augmentation, thanks to the richness and variety of the set. The YOLOv8n model, trained in Google Colab for 50 epochs, achieved an overall accuracy of 96% and an mAP50 greater than 94%, demonstrating high effectiveness under variable lighting environments. The YOLOv8 architecture enables real-time detection with low computational requirements, making it viable for applications in open field or packing plants. Despite its effectiveness, false positives were observed in the background class, suggesting the integration of techniques such as semantic segmentation as a future improvement. The proposed system contributes to agricultural automation, promoting more efficient, sustainable, and technologically advanced processes.