Early detection of clubroot in canola using drone-based hyperspectral imaging and machine learning

Abstract

Clubroot (Plasmodiophora brassicae) is spreading rapidly on canola (Brassica napus) in Canada. The disease often occurs first in small patches and then spreads across the field if not recognized and treated. Early detection is challenging because above-ground symptoms develop after the crop starts to flower, when scouting is difficult. Clubroot interferes with water uptake and delays flowering, which may result in changes in spectral reflectance that could be detected using a hyperspectral camera. The objective was to determine if a drone-mounted hyperspectral camera could be used to identify patches of clubroot from the air. Twenty-three research and commercial canola fields were imaged in Alberta and Saskatchewan during flowering from 2021 to 2023, using a remotely piloted aircraft system outfitted with a hyperspectral camera. One research site in Alberta offered an ideal mix of infected and non-infected canola for training a predictive classification model. Model development using machine learning (ML) and detailed plot mapping yielded the best results. Stochastic Gradient Boosting (SGB) consistently outperformed other ML classification algorithms tested. A 31-spectral band SGB model was subsequently used to assess 21 images from locations where comparisons with field sampling could be made with certainty. These comparisons yielded 100 % agreement in clubroot detection at the field level and > 90 % agreement for individual patches. Near infrared bands 758–764 nm were most important, especially 760 and 764 nm. Use of drones and hyperspectral technology offers promise for improved detection of clubroot so growers could choose appropriate crop rotations or treat infested patches.

Code and data availability

公開状態または取得可能な本文経路を確認できませんでした。

No evidence-backed public reproduction asset is currently recorded.