Development of a high-throughput plant disease symptom severity assessment tool using machine learning image analysis and integrated geolocation

Authors: CLOHESSY, JAMES - University Of Florida; SANJEL, SANTOSH - University Of Florida; O'BRIAN, G. KELLY - University Of Florida; BAROCCO, REBECCA - University Of Florida; KUMAR, SHIVENDRA - University Of Florida; Adkins, Scott; TILLMAN, BARRY - University Of Florida; WRIGHT, DAVID - University Of Florida; SMALL, IAN - University Of Florida

Submitted to: Computers and Electronics in Agriculture
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 3/4/2021
Publication Date: 3/31/2021
Citation: Clohessy, J.W., Sanjel, S., O'Brian, G., Barocco, R., Kumar, S., Adkins, S.T., Tillman, B., Wright, D.L., Small, I.M. 2021. Development of a high-throughput plant disease symptom severity assessment tool using machine learning image analysis and integrated geolocation. Computers and Electronics in Agriculture. 134:106089. https://doi.org/10.1016/j.compag.2021.106089.
DOI: https://doi.org/10.1016/j.compag.2021.106089

Interpretive Summary: Tomato spotted wilt virus (TSWV) can cause severe yield losses in peanut worldwide. Resistant peanut varieties are an effective means to manage TSWV. Breeding for TSWV resistance is difficult because accurate field-based assessment of disease incidence and severity is technically challenging and time-consuming. To address this challenge, a field-based, high-throughput assessment tool was developed to quantify the spatial distribution of disease symptoms over experimental peanut plots. Results from the disease assessment tool were compared with results from visual disease assessments, conducted by a trained plant pathologist. The results of this study demonstrate the successful application of this tool for high-throughput disease severity assessment in peanut under field conditions.

Technical Abstract: Tomato spotted wilt virus (TSWV) has the potential to cause severe yield losses in peanut, an important annual legume grown around the world. The most effective approach to manage the disease caused by TSWV is to grow disease resistant peanut varieties. One of the key challenges to breeding for disease resistance is to develop an accurate, reproducible and efficient disease assessment method. Consistent assessment across locations and seasons enables rapid genetic gain and reduces the cycles of selection required. Accurate field-based assessment of disease incidence and severity is technically challenging and time-consuming. To address this challenge, a field-based, high-throughput assessment tool was developed to quantify the spatial distribution of disease symptoms over experimental peanut plots using a Real Time Kinematic Global Positioning System (RTK-GPS), consumer-grade mirrorless cameras, a Raspberry Pi 3 microcontroller, and an open-source machine learning software. A field experiment was designed to establish a range of disease incidence and severity scenarios. This field experiment was imaged for two seasons to develop and validate the tool. Results from the disease assessment tool were compared with results from visual disease assessments, conducted by a trained plant pathologist. The results of this study demonstrate the successful application of this tool for high-throughput disease severity assessment in peanut under field conditions.