Sistema de estimación de producción de cultivo de mora empleando visión artificial y machine learning

Abstract

Blackberry cultivation is a significant sector of the national economy due to its high demand at both local and international levels. However, growers face considerable challenges in attempting to accurately predict production. Conventional methods, which rely heavily on manual procedures, are often inaccurate and prone to human error, potentially compromising planning and resulting in economic losses. The system developed in this research is composed of four fundamental stages: acquisition, processing, training, and analysis. For data collection, a DJI Mini 2 unmanned aerial vehicle (UAV) was utilized, capable of capturing real-time images of the crops. During processing, these images were analyzed using computer vision techniques, employing tools such as OpenCV and the YOLOv8m detection model. The model was trained with a specific dataset that included photographs of blackberries at different stages of maturity: green, red, and purple. To project potential losses due to factors such as pests, adverse weather, and drought, a Monte Carlo-based simulation was integrated. The results were presented in a graphical interface designed to facilitate visualization and analysis, assisting growers in optimizing their decision-making. The system exhibited 80.55% reliability in blackberry identification and counting, with data captured directly in the field. Furthermore, the simulations yielded a comprehensive evaluation of potential adverse scenarios, enabling more precise and realistic estimates.