Comprehensive Plant Health Monitoring: Expert-level Assessment with Spatio-Temporal Image Data

Alvaro Fuentes^1,2Syed Ali Asgher¹Jiuqing Dong¹Yongchae Jeong¹Mun Haeng Lee³Taehyun Kim⁴Sook Yoon⁵Dong Sun Park^1,2*

• ¹Department of Electronics Engineering, Jeonbuk National University, Jeonju, North Jeolla, Republic of Korea
• ²Core Research Institute of Intelligent Robots, Jeonbuk National University, Jeonbuk National University, Republic of Korea
• ³Department of Smart Farm, Chungnam State University, Chungcheongnam, Republic of Korea
• ⁴National Institute of Agricultural Sciences, Wanju, Republic of Korea
• ⁵Department of Computer Engineering, College of Engineering, Mokpo National University, Muan, Republic of Korea

Maintaining crop health is critical for global food security, as plants are increasingly affected by pathogens and environmental stressors. However, conventional plant health monitoring relies heavily on manual inspection, making it labor-intensive and insufficient for timely, proactive interventions. In this study, we propose a deep learning-based framework for expert-level, spatio-temporal plant health assessment using sequential RGB images. Unlike conventional models that focus on isolated disease detection, our approach enables continuous health monitoring throughout the cultivation cycle, categorizing plant health into five levels-from very poor to optimal-based on visual and morphological indicators. To validate our method, we collected a custom dataset of 12,119 annotated images from 200 tomato plants of three varieties grown in semi-open greenhouses across multiple seasons. The framework leverages state-of-the-art CNN and transformer architectures to generate accurate, stage-specific health predictions, which align closely with expert annotations. These predictions enable the creation of cultivation maps over time, facilitating early intervention and datadriven crop management. The results demonstrate the framework's effectiveness for precision agriculture and scalable, real-world deployment.