Real-time tomato leaf disease detection and severity assessment on embedded AI platforms

Authors: NARASIMHAMURTHY, KRUTHIK - University Of Texas At Tyler; SUNDARAVADIVEL, PRABHA - University Of Texas At Tyler; TAMIL, LAKSHMAN - University Of Texas At Tyler; Ling, Kai Shu

Submitted to: Meeting Abstract
Publication Type: Abstract Only
Publication Acceptance Date: 3/5/2026
Publication Date: 4/30/2026
Citation: Narasimhamurthy, K., Sundaravadivel, P., Tamil, L., Ling, K. 2026. Real-time tomato leaf disease detection and severity assessment on embedded AI platforms. Meeting Abstract. https://units.cals.ncsu.edu/2026-ai-ag-conference/.

Interpretive Summary: N/A

Technical Abstract: Timely and accurate identification of plant diseases is essential for effective crop management and yield protection, particularly in field environments where delays in diagnosis can lead to rapid disease spread and significant economic losses. Tomato crops are especially vulnerable to viral diseases such as tomato yellow leaf curl virus and tomato brown rugose fruit virus, whose symptoms can vary across growth stages and environmental conditions. Many existing plant disease detection solutions depend on cloud-based processing or offline analysis, which introduces latency, connectivity constraints, and practical deployment challenges in real-world agricultural settings. The purpose of this work is to develop and evaluate an edge-deployed computer vision system capable of real-time detection and severity assessment of tomato diseases directly at the point of capture, enabling stage-aware and actionable disease monitoring for precision agriculture applications. This study utilizes multiple tomato leaf image datasets containing representative samples of tomato yellow leaf curl virus and tomato brown rugose fruit virus to analyze disease characteristics and symptom progression across different stages of infection. A convolutional object detection model is trained to localize diseased leaf regions and distinguish infection patterns from healthy tissue under varying lighting and background conditions. Rather than limiting analysis to binary healthy–diseased classification, the approach incorporates visual indicators such as leaf curl intensity, discoloration patterns, and the spatial distribution of affected regions to support severity estimation. To enable real-time, in-field operation, the trained model is deployed on NVIDIA Jetson Orin embedded platforms. This edge deployment allows inference to be performed directly on live camera streams without reliance on cloud connectivity. The system architecture is designed to operate within the computational and power constraints typical of embedded agricultural sensing platforms, supporting low-latency processing while maintaining detection accuracy. Performance is evaluated under realistic conditions to assess detection reliability, responsiveness, and suitability for continuous field monitoring. Experimental evaluation demonstrates that the edge-deployed computer vision system achieves reliable disease detection performance across varying environmental conditions, including changes in illumination, leaf orientation, and background complexity. The model effectively localizes infected regions and differentiates disease symptoms from healthy foliage while operating within the power and compute limits of the embedded hardware. Severity assessment based on observable visual features enables stage-aware characterization of disease progression, providing richer information than binary classification approaches. The results confirm that real-time inference on embedded platforms is feasible and effective for tomato disease monitoring without dependence on external cloud infrastructure. By performing disease detection and severity assessment directly at the edge, this work addresses key limitations of cloud-dependent plant disease monitoring systems, including latency, connectivity requirements, and deployment complexity. The proposed framework enables timely, actionable insights that can support early intervention, targeted treatment, and improved crop management decisions. Its stage-aware design aligns more closely with practical agricultural workflows, where understanding disease progression is critical for selecting appropriate responses. The demonstrated edge-based approach offers a scalable and field-ready solution for precision agriculture and establishes a foundation for extending embedded AI systems to other crops and disease types, advancing real-world adoption of intelligent plant healt