Robust real-time strawberry maturity detection using UAV-mounted deep learning for precision agriculture.

BACKGROUND: To address the challenge of real-time plant monitoring in greenhouse environments, this industry-driven research focuses on developing an autonomous quadrotor UAV system specifically designed for monitoring strawberry plants. Traditional methods for greenhouse monitoring are labor-intensive and lack scalability, particularly in precision agriculture applications. METHOD: This research introduces a mature strawberry detection model specifically designed for greenhouse environments. The proposed YOLOv9-GLEAN approach enables the identification of small mature strawberries through an onboard camera mounted on the quadrotor. Additionally, a hybrid trajectory tracking controller for the quadrotor is developed and tested in both simulated and real-world conditions. The UAV navigates through the greenhouse using predetermined waypoints, operating as a semi-autonomous system for navigation while maintaining full autonomy in mature strawberry detection tasks. The system incorporates an integrated onboard vision platform that utilizes an innovative YOLOv9-GLEAN-based algorithm to perform real-time and offline detection and counting of mature strawberries. RESULTS: The YOLOv9-GLEAN model achieves high detection accuracy, as confirmed by evaluation metrics such as precision, recall, and F1-score. The proposed hybrid (PID+LQR) controller demonstrates superior tracking performance compared to other conventional controllers. The integrated control and perception system proves effective in both simulated and real-world greenhouse environments. DISCUSSION: The research validates the efficacy of deep learning models, with YOLOv9-GLEAN showing exceptional performance in enabling rapid, precise, and automated detection of ripe strawberries through quadrotor deployment in greenhouse environments. Such agricultural monitoring technologies represent a substantial advancement beyond conventional manual inspection approaches, empowering farmers and greenhouse operators to execute well-informed, time-sensitive management decisions that minimize crop losses and optimize production yields. This investigation underscores the revolutionary impact that deep learning technologies can have within greenhouse agriculture.