Wheat Fusarium Head Blight Disease Severity Estimation using UAS Multispectral Imagery and Machine Learning

Le résumé fourni par la source

Wheat is an important primary crop that feeds billions of people worldwide. Wheat diseases, particularly Fusarium Head Blight (FHB), often have severe effects on yield quality and the health of humans and livestock. Traditional field-based methods for monitoring wheat diseases are time-consuming and inefficient. Remote sensing approach, particularly the use of aerial imaging via Uncrewed Aircraft Systems (UAS) has become an essential tool for fine-scale and rapid field scouting and crop disease monitoring in recent years. This study investigates the potential of combining high-resolution UAS multispectral imagery with machine learning (ML) methods to detect FHB disease severity. Two experimental wheat fields were set up in Brookings, South Dakota, USA, in 2022. The severity of FHB disease was assessed periodically through visual observation, with synchronous UAS flights collecting multispectral imagery. UAS-based canopy spectral and texture features were derived and used as input variables to develop ML-based classification and regression models to estimate FHB disease severity. Conventional ML methods such as Support Vector Machine (SVM) and Random Forest (RF), along with deep learning models like Deep Neural Network (DNN) and One-Dimensional Convolutional Neural Network (1D-CNN), were employed. SVM demonstrated superior performance in both classification and regression analyses for predicting FHB disease severity levels, achieving an overall accuracy of 0.80 using texture features in classification and an R2 of 0.73 using spectral features in regression. The results show that both spectral and texture features are important features for estimating wheat FHB disease severity. UAS remote sensing, coupled with ML-based approaches, provides a rapid and accurate method to assess FHB disease severity at a fine scale over large areas.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.