Soil Nutrient Estimation and Mapping in Farmland Based on UAV Imaging Spectrometry

Area of Science:

• Agricultural Science
• Remote Sensing
• Soil Science

Background:

• Soil nutrient status is crucial for farmland productivity.
• Imaging spectrometry offers rapid, real-time soil characteristic monitoring.
• Hyperspectral imaging from UAVs presents a promising tool for soil analysis.

Purpose of the Study:

• To explore preprocessing and modeling methods for UAV-based hyperspectral image analysis.
• To estimate soil organic matter (SOM) and soil total nitrogen (STN) in farmland.
• To enhance the accuracy of soil nutrient estimation using advanced algorithms.

Main Methods:

• Multiplicative Scattering Correction (MSC) was used for spectral noise reduction.
• A combined Successive Projections Algorithm (SPA) and Competitive Adaptive Reweighted Sampling (CARS) method selected optimal hyperspectral bands.
• Particle Swarm Optimization (PSO) optimized an Extreme Learning Machine (ELM) model for prediction.

Main Results:

• MSC outperformed SNV and spectral derivatives in reducing noise and improving signal.
• 24 and 22 feature bands were selected for SOM and STN estimation, respectively.
• The PSO-ELM model achieved higher prediction accuracy (R²=0.73 for SOM, R²=0.63 for STN) compared to SVM, PLSR, and basic ELM.

Conclusions:

• Optimized hyperspectral data preprocessing and feature selection are vital for accurate soil nutrient estimation.
• The PSO-ELM model demonstrates superior performance for predicting SOM and STN.
• This study provides a valuable framework for precision agriculture using imaging spectrometry.