Deriving Early Citrus Fruit Yield Estimation by Combining Multiple Growing Period Data and Improved YOLOv8 Modeling.

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Early crop yield prediction is vital for precision agriculture and sustainable fruit production.
• Accurate detection of fruit developmental stages (flowering, green fruiting, ripening) is crucial for yield estimation.
• Existing YOLO model studies often focus on single maturity stages, limiting comprehensive analysis.

Purpose of the Study:

• To develop an efficient and rapid crop yield estimation model for sustainable fruit production.
• To propose a novel lightweight network, YOLOv8-RL, for analyzing citrus multigrowth period characteristics.
• To construct and validate a citrus yield estimation model using the proposed network.

Main Methods:

• Proposed the YOLOv8-RL network model utilizing citrus multigrowth period data.
• Constructed and validated a citrus yield estimation model by integrating network identification counts with manual field counts.
• Evaluated model performance based on recognition rates, mAP@.5, FPS, inference time, and prediction error rates.

Main Results:

• The YOLOv8-RL model demonstrated a 50.7% reduction in parameters and a 49.4% decrease in floating-point operations compared to YOLOv8, with a model size of only 3.2 MB.
• Achieved an average recognition rate of 95.6% for citrus flowers, green fruits, and orange fruits, with an mAP@.5 of 94.6% and high FPS (123.1).
• Yield estimation models showed high accuracy with coefficients of determination (R²) of 0.91992 and 0.95639, and low prediction error rates (6.96% for green fruits, 3.71% for orange fruits).

Conclusions:

• The YOLOv8-RL model offers a lightweight and efficient solution for crop yield prediction, suitable for embedded devices.
• The developed yield estimation models provide accurate and reliable predictions, outperforming traditional network counting methods.
• This study offers a theoretical basis and technical support for early fruit yield prediction in complex agricultural environments.