Bell pepper detection system using YOLOv8n

Traditional crop monitoring in Philippine agriculture heavily relies on manual inspection. This process is highly susceptible to human error, labor-intensive, and prone to causing significant financial losses due to inconsistent ripeness assessment. While computer vision and deep learning offer promising non-contact solutions for precision agriculture, many existing object detection frameworks require substantial computational resources, limiting their deployment on portable, field-ready hardware. This study addresses these challenges by developing a lightweight, real-time detection and classification system for local bell peppers (Sultan F1) utilizing the YOLOv8n variant architecture. Following an Agile development methodology—encompassing system requirements, design, testing, and deployment—the system was trained on an annotated dataset of raw images captured in an indoor environment, categorized into "monitor" and "ripe" maturity stages. The developed system integrates computer vision with a user-friendly monitoring dashboard to display real-time classification and ripeness outputs, optimizing inference speed and resource efficiency. The results demonstrate the viability of implementing edge AI on low-resource, portable platforms to mitigate post-harvest losses. Ultimately, this lightweight architecture provides a foundational framework for future advancements in smart harvesting, autonomous agricultural robotics, and automated crop monitoring systems. Both the training and evaluation results showed that YOLOv8n achieved nearly perfect performance across metrics such as F1-score, precision, recall, and mean Average Precision (mAP). The confidence curves of the metrics also verified the model’s accuracy in detecting the bell pepper (Sultan F1) in terms of its maturity level: green pepper and red pepper (ripe) or transition pepper (monitor).