Improved leaf area index estimation in Jujube trees by fusing spatial-temporal features from UAV RGB time-series with...

Area of Science:

• Agricultural Remote Sensing
• Deep Learning Applications
• Plant Physiology

Background:

• Traditional vegetation indices for Leaf Area Index (LAI) estimation face saturation and background noise issues.
• Accurate LAI estimation is crucial for precision agriculture and smart orchard management.
• Drone remote sensing offers high-resolution data but requires advanced analytical methods.

Purpose of the Study:

• To develop a high-precision, high-efficiency Leaf Area Index (LAI) estimation method for jujube trees using drone remote sensing.
• To overcome the limitations of traditional vegetation index methods.
• To propose a parallel hybrid deep learning framework integrating spatial and temporal features.

Main Methods:

• A parallel hybrid deep learning framework combining Convolutional Neural Network (CNN) for spatial-spectral feature extraction and Gated Recurrent Unit (GRU) for temporal dynamics was developed.
• Multi-temporal ground-measured LAI data and synchronized drone RGB images were collected over two years.
• Spectral and texture indices were extracted as inputs, and a meta-learner integrated spatial-temporal information.

Main Results:

• The CNN-GRU model demonstrated strong performance on the training set (R² = 0.839).
• Optimized with data augmentation, the model achieved high prediction accuracy on the test set (R² = 0.83, RMSE = 0.150).
• The proposed model outperformed mainstream comparative models like Transformer, KNN, MLP, and CNN.

Conclusions:

• The hybrid deep learning architecture effectively integrates spatial and temporal information for robust crop LAI remote sensing inversion.
• This approach provides a reliable technical tool for digital management in smart orchards and precise agricultural decision-making.
• The CNN-GRU framework offers a significant advancement over traditional methods for complex agricultural scenarios.