Predicting F _v /F _m and evaluating cotton drought tolerance using hyperspectral and 1D-CNN

Area of Science:

• Plant Physiology
• Agricultural Engineering
• Remote Sensing

Background:

• Chlorophyll fluorescence (F) is crucial for assessing plant abiotic stress.
• Current F measurement methods require dark adaptation, limiting real-time, high-throughput analysis.
• Hyperspectral imaging shows potential for analyzing large, diverse plant samples.

Purpose of the Study:

• To explore hyperspectral prediction of chlorophyll fluorescence (Fv/Fm) in cotton for drought tolerance evaluation.
• To develop and validate models for rapid, non-destructive assessment of cotton drought status.
• To identify superior drought-tolerant cotton genotypes using advanced analytical techniques.

Main Methods:

• Studied 80 cotton varieties under normal and drought stress conditions across key growth stages.
• Acquired hyperspectral data and applied various machine learning models, including 1D-CNN, CatBoost, and Random Forests, to predict Fv/Fm.
• Utilized the Savitzky-Golay filter combined with 1D-CNN for optimal model performance.

Main Results:

• The Savitzky-Golay + 1D-CNN model demonstrated superior accuracy and robustness in predicting Fv/Fm (RMSE = 0.016, MAE = 0.009).
• Predicted Fv/Fm drought tolerance coefficients closely matched manually measured values.
• The model successfully differentiated cotton varieties based on drought tolerance.

Conclusions:

• Hyperspectral full-band technology and 1D-CNN modeling offer a non-destructive, fast, and accurate method for monitoring cotton drought status.
• This approach facilitates the identification of drought-tolerant cotton genotypes, supporting smart agriculture.
• The study validates the potential of hyperspectral Fv/Fm prediction for large-scale crop management.