Avocado Detection System on a UAV with YOLOv8 Image Processing

Yanyachi P.R.; Aco A.E.L. · IEEE Access (2026) · DOI: 10.1109/access.2026.3687453

Resumen

The automation of agriculture increasingly relies on Unmanned Aerial Vehicles (UAVs); however, the detection and analysis of fruits such as Hass avocados remain challenging due to their irregular shape, dark skin, and complex background. This work proposes a computer vision system for avocado detection and analysis that integrates a YOLOv8 (You Only Look Once version 8) neural network to perform maturity classification, fruit counting, distance estimation, height calculation, and 2D mapping within a fixed reference frame defined by the UAV. The system was evaluated in both simulated and real environments. In the simulated environment, the entire processing pipeline was validated, achieving reliable detection and stable spatial analysis performance. In real-world tests, only the image acquisition stage was assessed using an onboard vision platform, demonstrating the feasibility of in-flight image capture for subsequent processing. Preliminary results from the training of multiple YOLOv8 variants showed consistent learning behavior, with the YOLOv8s model achieving the best balance between accuracy and efficiency. It reached a mean Average Precision at IoU = 0.5 (mAP@0.5) of 84.01% and a mean Average Precision at IoU range = 0.5:0.95 ( mAP@0.5:0.95) of 76.94%, along with a precision of 79.66% and a recall of 85.37%. These values confirm the model's robust detection capability and validate the reliability of the proposed system in both simulated and real environments, highlighting its potential for future real-Time UAV-based fruit detection in precision agriculture.

Instituciones: Universidad Nacional de San Agustín de Arequipa