Faba bean and pea harvest index estimations using aerial-based multimodal data and machine learning algorithms

Area of Science:

• Agricultural Science
• Remote Sensing
• Machine Learning

Background:

• Traditional crop harvest index (HI) estimation is labor-intensive and time-consuming.
• Unmanned aerial vehicles (UAVs) offer a potential solution for efficient HI assessment.
• Multimodal sensor data fusion can enhance remote sensing capabilities in agriculture.

Purpose of the Study:

• To explore the use of low-cost, UAV-based multimodal data (RGB, MS, TIR) for estimating faba bean and pea harvest index (HI).
• To evaluate the effectiveness of ensemble learning, specifically Bayesian model averaging, for HI estimation.
• To assess the impact of different growth stages on HI estimation accuracy.

Main Methods:

• Collected UAV-based RGB, multispectral (MS), and thermal infrared (TIR) data across four growth stages.
• Employed ensemble learning, including Bayesian model averaging, for HI estimation.
• Compared the performance of individual sensors and data fusion techniques.

Main Results:

• Multisensor data fusion significantly improved HI estimation accuracy compared to individual sensors.
• Ensemble Bayesian model averaging achieved the highest accuracy (faba bean: R2 = 0.64, NRMSE = 13.76%; pea: R2 = 0.74, NRMSE = 15.20%).
• Estimation accuracy increased with crop growth stage.

Conclusions:

• Combining low-cost, UAV-based multimodal data with machine learning provides a reliable method for crop HI estimation.
• This approach supports high spatial precision in agriculture, enabling early and efficient decision-making for breeders and field managers.
• The study highlights a promising strategy for advancing precision agriculture techniques.