Inversion of cotton leaf area index under Verticillium wilt stress from UAV multispectral images: Deep learning-based vs. classical-based algorithms

Abstract

A precise and timely assessment of the cotton leaf area index (LAI) plays a critical role in advancing precision agriculture practices. Verticillium wilt reduces LAI by disrupting leaf structure and physiological functions.Verticillium wilt is a significant disease that constrains the yield potential of cotton. The emergence of unmanned aerial vehicle (UAV) technology offers a revolutionary technology for cost-effective, fine-scale inversion of the LAI in cotton under Verticillium wilt stress. The study investigated the relevant potential of classical-based algorithms and deep learning-based algorithms to estimate the LAI using multi-spectral images acquired by drones under Verticillium wilt stress. Prior to constructing the model, Gaussian blur was applied to enhance the original images. Inputs to the Random Forest (RF) and Extreme Learning Machine (ELM) algorithms included vegetation indices (VI) and texture features (TF), and the LAI of cotton under Verticillium wilt stress was estimated in conjunction with ground-measured LAI data. The enhanced images were used as input to the Convolutional Neural Network (CNN) algorithm to estimate the LAI of cotton under Verticillium wilt stress. The results demonstrated that, in the classical-based models, the prediction accuracy of LAI inversion using both VI and TF was slightly higher than that achieved using VI or TF alone. After Gaussian blur processing, the accuracy of RF, ELM, and CNN models showed significant improvement. Among these, the CNN model based on Gaussian blur achieved the highest LAI inversion accuracy (R²≥0.88), followed by the LSM model (R²=0.83) and the RF model (R²=0.82). Consequently, the CNN model under Gaussian blur is employed to achieve rapid and accurate inversion of the LAI of cotton under Verticillium wilt stress. This finding of the study provides a significant reference for real-time monitoring of cotton growth and effective field management under Verticillium wilt conditions.

Topics & Concepts