Prediction of strawberry yield based on receptacle detection and Bayesian inference.

Area of Science:

• Agricultural Science
• Computer Vision
• Statistical Modeling

Background:

• Strawberry yield prediction is crucial for agricultural management.
• Receptacles are direct precursors to fruit, making them ideal for yield forecasting.
• Existing methods may not fully capture the complexities of yield prediction.

Purpose of the Study:

• To develop and evaluate an AI-driven method for predicting strawberry yield.
• To combine object detection for receptacle counting with Bayesian analysis for yield forecasting.
• To compare the efficacy of receptacle-based prediction with traditional early-season yield monitoring.

Main Methods:

• Utilized Faster R-CNN for object detection to count strawberry receptacles in images.
• Employed Bayesian analysis to model uncertainty and account for missed receptacles.
• Integrated receptacle counts and fruiting probabilities into yield prediction models.
• Collected data over two years across five major cultivars considering environmental variations.

Main Results:

• Achieved a mean Average Precision (mAP) of 0.6587 for receptacle detection using Faster R-CNN.
• Demonstrated that receptacle counting via object detection, combined with Bayesian modeling, significantly improves yield prediction accuracy.
• Found this AI-assisted approach more effective than using cumulative yield data from the first month.

Conclusions:

• AI-based receptacle detection and Bayesian modeling offer a robust strategy for strawberry yield prediction.
• This approach provides a more accurate forecast of total yield compared to early-season cumulative data.
• The study highlights the potential of integrating advanced computational techniques in precision agriculture.