Feature Selection and Spectral Indices for Identifying Maize Stress Types.

Area of Science:

• Agricultural remote sensing
• Plant physiology
• Spectroscopy

Background:

• Accurate monitoring of maize stress is crucial for agricultural productivity.
• Spectral analysis offers a non-destructive method for plant health assessment.
• Differentiating various stress types (heavy metal, water, fertilizer) is challenging but vital.

Purpose of the Study:

• To identify spectral features and indices for distinguishing different maize leaf stress types.
• To evaluate the performance of various feature selection and classification algorithms.
• To provide a technical basis for maize stress monitoring.

Main Methods:

• Spectral data preprocessing including continuum removal (CR) and standard normal variable (SNV) transformation.
• Application of feature selection methods (ReliefF, RF, GBT) to identify sensitive spectral bands.
• Construction of spectral indices and classification using algorithms like Random Forest (RF) and Adaptive Boosting (AdaBoost).

Main Results:

• Key spectral bands for stress differentiation are located in the red edge (700-800 nm) and water absorption valley (1900 nm).
• Spectral indices utilizing near-infrared plateau (1185 nm) and water absorption (1460 nm) bands effectively differentiated stress types.
• Random Forest (RF) and Adaptive Boosting (AdaBoost) demonstrated optimal classification performance.

Conclusions:

• Spectral analysis combined with advanced machine learning can accurately identify and differentiate maize stress types.
• The study provides valuable insights into spectral characteristics associated with specific plant stresses.
• Findings support the development of remote sensing tools for precision agriculture and crop management.