Battle Royale Optimization for Optimal Band Selection in Predicting Soil Nutrients Using Visible and Near-Infrared...

Area of Science:

• Agricultural Science
• Soil Science
• Remote Sensing

Background:

• Soil property quantification is crucial for precision agriculture.
• Traditional methods are labor-intensive and time-consuming.
• Hyperspectral remote sensing offers a rapid, non-destructive approach.

Purpose of the Study:

• To quantify soil properties (pH, EC, organic carbon, N, P, K) using hyperspectral remote sensing.
• To evaluate the performance of various metaheuristic algorithms for optimizing spectral bands.
• To determine the most effective algorithm for predicting soil nutrients.

Main Methods:

• Collected 100 soil samples with diverse properties.
• Measured spectral reflectance (350-1050 nm) using SVC GER 1500 Spectroradiometer.
• Employed Particle Swarm Optimization, Moth-Flame Optimization, Flower Pollination Optimization, and Battle Royale Optimization for band selection.
• Utilized Partial Least Squares Regression (PLSR) for prediction modeling.

Main Results:

• Battle Royale Optimization (BRO) demonstrated the best performance.
• R2 values for BRO were 0.45 (pH), 0.32 (EC), 0.48 (organic carbon), 0.21 (AN), 0.71 (AP), and 0.35 (AK).
• Quantification of soil nutrients from spectral data achieved fair to good accuracy.

Conclusions:

• Hyperspectral remote sensing combined with machine learning, particularly BRO, is effective for soil property quantification.
• This approach offers a promising alternative to traditional soil analysis methods.
• Further research can optimize algorithms and expand the range of quantifiable soil properties.