Thermal Imaging and Disease Detection in Hibiscus Plant

Abstract: The threat of plant diseases poses a significant challenge to agricultural productivity, especially in developing countries where small-scale farmers are highly vulnerable. To avoid crop loss and guarantee food security, early identification of plant stress and disease is crucial. Even if they work well, traditional diagnostic techniques take a lot of time and effort. This study explores the potential of thermal imaging as a non-invasive and efficient solution for early stress detection in Hibiscus plants. The experiment was conducted on a single potted hibiscus plant at Bikaner Technical University over a period of approximately two months (14th December 2024 to 5th February 2025). Initially kept outdoors with regular watering, the plant was moved to a closed indoor setting without sunlight and water from Day 9, allowing for observation of stress progression and disease emergence. Although the thermal dataset was recorded for 16 days, intermediate day observations confirmed consistent stress behavior. Four visual diseases—Leaf Spot, Rust Disease, Botrytis Blight, and Mosaic Virus—were noted, but due to the limited dataset, the prototype focuses on classifying plant health into four thermal stress categories: healthy, mild, significant, and critical. Thermal images were captured from top and front views, and a deep learning model based on MobileNetV2 was developed using a multi-view classification approach. The model was trained using Leave-One-Out Cross-Validation (LOOCV) to ensure robustness with constrained data. Instead of relying solely on traditional performance metrics, a confidence-based interpretation method was adopted to improve decision reliability. The prototype demonstrates the feasibility of using thermal imaging and deep learning for early, non-destructive plant stress classification, paving the way for smarter and more sustainable agricultural monitoring.