Click on “Download PDF” for the PDF version or on the title for the HTML version.


If you are not an ASABE member or if your employer has not arranged for access to the full-text, Click here for options.

Using aerial imagery coupled with machine learning to assess Goss's Wilt disease severity in field corn

Published by the American Society of Agricultural and Biological Engineers, St. Joseph, Michigan www.asabe.org

Citation:  2021 ASABE Annual International Virtual Meeting  2100146.(doi:10.13031/aim.202100146)
Authors:   Anup Kumar Das, Andrew Friskop, Paulo Flores, Cannayen Igathinathane, Jithin Jose Mathew, Zhao Zhang
Keywords:   Goss’s Wilt, Severity Assessment, Corn, Image Processing, Machine Learning

Abstract. Goss's Wilt has been identified as one of the most yield-limiting diseases in corn production in North Dakota (ND), with reported yield losses of up to 50%. Chemical applications (e.g., hydroxides and citric acid) at the proper corn growth stage can minimize or even avoid yield losses. A key aspect for chemical applications is to assess the level of infestation and the severity of the disease across a field. The conventional assessment is completed through field visits and visual observations, which can be time consuming (inefficient), subjective, and can lead to inaccuracy due to evaluator‘s fatigue. Current technological advancement of unmanned aerial systems (UASs) incorporated with machine learning (ML) techniques provides a new approach for disease severity assessment. However, the correlation between ML and human-based assessment has not been well investigated. In this study, we collected UAS imagery presenting different severity levels from the same corn fields. Shortly after the UAS flights, plant pathologists surveyed the field plots and visually rated the Goss‘s Wilt severity of each plot. A dataset containing 200 images was prepared from the UAS imagery and then fed into different ML algorithms. To be specific, different textural, color, and area features were extracted and fed into different ML algorithms, followed by the performance of different ML algorithms compared. Random forest (RF) classifier achieved highest precision (72.30%), recall (72.50%) and F-score (72.39%) than other classifiers including Naive bayes, Support vector machine, Feedforward artificial neural network, Logistic regression, Stochastic gradient descent, and Ada boost. Hue value and infected leaves area are important factors for identifying Goss Wilt severity. Therefore, the developed automatic Goss‘s Wilt disease severity detection based on UAS imagery coupled with RF classifier can be a potential tool to aid farmers in monitoring Goss‘s Wilt disease.

(Download PDF)    (Export to EndNotes)