Rice LAI estimation using UAV-based multi-parameter fusion at the booting stage.

Area of Science:

• Agricultural Science
• Remote Sensing
• Machine Learning

Background:

• Leaf Area Index (LAI) is vital for crop growth and water use efficiency.
• Traditional LAI measurement methods are time-consuming, labor-intensive, and difficult to scale.
• Accurate LAI monitoring is essential for effective precision rice management.

Purpose of the Study:

• To develop an efficient and accurate framework for estimating rice LAI using multi-source UAV remote sensing data.
• To overcome the limitations of traditional LAI measurement techniques.
• To optimize LAI estimation models for the rice booting stage through machine learning.

Main Methods:

• Integrated multi-source UAV remote sensing features: color indices (CIs) from RGB, vegetation indices (VIs) from multispectral data, texture features (TIs), and texture feature indices (TFIs).
• Employed six machine learning algorithms to develop optimized LAI estimation models.
• Evaluated models at different flight altitudes (30m and 60m).

Main Results:

• At 30m, CNN integrating CIs and TIs achieved R²=0.815.
• At 60m, RF combining VIs and TFIs achieved R²=0.866.
• The best performance (R²=0.901, RMSE=0.273, RPD>3.0) was obtained by integrating CIs, VIs, and TFIs at 30m, demonstrating the effectiveness of TFIs and multi-source data fusion.

Conclusions:

• Texture feature indices (TFIs) significantly enhance multispectral data's spectral-spatial representation, improving model accuracy.
• Combining color indices (CIs) and vegetation indices (VIs) compensates for spectral and spatial limitations.
• The integrated multi-source, multi-resolution approach provides a robust solution for high-precision LAI estimation in rice, supporting precision agriculture.