Digital tool for counting coffee plants and economic study of alternative crops

Traditional methods for counting coffee plants in large-scale agricultural plantations rely on manual processes, which are time-consuming and prone to errors. This paper presents a method for automated coffee plant counting using aerial imagery captured by drones and processed with the YOLOv8 (You Only Look Once) deep learning model. The proposed solution achieves an mAP50-95 of 67.54% and a precision of 87.50%, effectively detecting coffee plants in real-world plantation scenarios. Challenges such as overlapping crops and visually similar vegetation were addressed during the creation of a manually labeled dataset for model training. An interactive platform facilitates model evaluation and inference. The method is robust in detecting mature plants and provides an efficient alternative to manual counting, significantly improving productivity at a reduced cost. However, there are opportunities to improve performance to detect young and partially hidden plants. This approach lays the foundation for further advancements in agricultural automation and the estimation of coffee production.