High-throughput method for improving rice AGB estimation based on UAV multi-source remote sensing image feature...

Area of Science:

• Agricultural Remote Sensing
• Plant Physiology
• Machine Learning

Background:

• Accurate estimation of rice aboveground biomass (AGB) is crucial for crop management and yield prediction.
• Traditional vegetation indices (VIs) struggle with saturation in dense canopies, limiting their effectiveness across rice growth stages.
• Unmanned Aerial Vehicle (UAV) imagery offers a promising, non-destructive approach for AGB assessment.

Purpose of the Study:

• To explore the potential of fusing UAV-acquired RGB and multi-spectral (MS) image data for accurate and cost-effective rice AGB estimation.
• To evaluate the performance of single-sensor features versus multi-source data fusion.
• To assess the efficacy of ensemble machine learning (ML) models for AGB prediction.

Main Methods:

• Extracted high-frequency texture features from RGB images using discrete wavelet transform (DWT) and calculated color moments.
• Derived vegetation indices (VIs) from MS images.
• Employed feature selection techniques, including Variance Inflation Factor (VIF) for collinearity removal, and developed individual and ensemble ML models.

Main Results:

• Multi-feature fusion significantly improved AGB estimation accuracy compared to single-sensor features.
• Fusing RGB and MS image features enhanced accuracy over using either sensor alone.
• Ensemble ML models demonstrated superior accuracy and stability, with the best model achieving R² = 0.8564 and RMSE = 169.32 g/m².

Conclusions:

• Multi-source UAV image feature fusion combined with ensemble learning provides an efficient and accurate solution for monitoring rice AGB.
• This approach effectively leverages complementary data strengths for robust crop biomass estimation.
• The study highlights the value of integrated remote sensing and ML techniques in precision agriculture.