ABSTRACT
Artificial intelligence-based detection of Cercospora leaf spot disease severity in mung bean

Pitchakon Papan1, Witsarut Chueakhunthod1, Chanwit Kaewkasi2, Wanploy Jinagool1, Akkawat Tharapreuksapong3, Arada Masari4, Sumana Ngampongsai4, Kanlayanee Sawangsalee1, and Piyada A. Tantasawat1*
 
Cercospora leaf spot (CLS), caused by Cercospora canescens Ellis and G. Martin, is a significant foliar pathogen impacting mung bean (Vigna radiata (L.) R. Wilczek) cultivation. Early and precise detection of plant diseases via advanced digital technologies is pivotal for optimizing disease management strategies and developing cultivars with enhanced resistance. In this study, we employed an artificial neural network (ANN) model to assess CLS in mung bean, utilizing a dataset acquired under field conditions, and benchmarked its performance against traditional visual rating methodologies. Our analysis incorporated a deep learning framework based on the EfficientNet-B3 architecture. This model demonstrated superior performance in detecting CLS symptoms from images, achieving a validation accuracy of 88.70% and exhibiting a reduced overfitting gap between training and validation datasets (4.54%), relative to the Inception V3 model, which achieved 82.21% accuracy and an overfitting gap of 12.47%. Furthermore, the CLS index derived from the EfficientNet-B3 model showed a strong correlation with expert-rated CLS scores in field conditions (r = 0.925**, R² = 0.856). The application prototype developed in this research accurately assessed CLS severity, aligning closely with visual evaluations by experts (r = 0.910**, R² = 0.828). These findings underscore the potential of integrating deep learning techniques into plant disease management, providing a reliable and efficient tool for assessing CLS severity, which can enhance decision-making processes in crop management and facilitate research advancements.
Keywords: Artificial intelligence, artificial neural network, Cercospora leaf spot, machine learning, Vigna radiata, visual observation.
1Suranaree University of Technology, Institute of Agricultural Technology, School of Crop Production Technology, 30000 Nakhon Ratchasima, Thailand.
2Suranaree University of Technology, Institute of Engineering, School of Computer Engineering, 30000 Nakhon Ratchasima, Thailand.
3Suranaree University of Technology, Center for Scientific and Technological Equipment, 30000 Nakhon Ratchasima, Thailand.
4Field and Renewable Energy Crops Research Institute, 10900, Bangkok, Thailand.
*Corresponding author (piyada@sut.ac.th).