Multi-Temporal, Multi-Modal UAV and Machine Learning Framework for Early Detection and Mapping of Bacterial Leaf Blight in Rice

Abstract. Bacterial leaf blight (BLB), caused by infections with Xanthomonas oryzae pv. Oryzae (Xoo) bacteria, is one of the most destructive diseases affecting rice cultivation across the globe, frequently leading to severe yield losses. Early detection and timely management before visible symptoms appear and irreversible damage occurs are crucial to mitigating its impact on rice productivity and ensuring food security. Traditional monitoring methods based on field inspections require expert knowledge, are limited to visible symptom stages, and are time-consuming. Recent advancements in remote sensing, particularly with uncrewed aerial vehicles (UAVs), offer new opportunities for timely, quantitative, high-resolution monitoring of crop conditions. This study aimed to develop a multi-temporal, multi-modal UAV-based framework integrating multispectral, thermal, and RGB data to detect and map early-stage BLB infections in rice canopies. During the 2023 wet season at the Zeigler Experiment Station of the International Rice Research Institute (IRRI), an experiment was established using two rice fields, each equally divided into healthy and artificially inoculated plots. Disease severity percentage at specific growth stages, with weekly UAV imagery, were collected throughout the growing season. UAV imagery was processed to derive spectral, thermal, and textural features, including GNDVI, NDRE, canopy temperature, and texture indices. A Random Forest machine learning algorithm was applied to identify the most sensitive indicators of early BLB infection. Our results showed that integrating spectral and thermal features enabled earlier-stage BLB detection. The study offers a deployable, data-driven framework for precision rice disease monitoring, enabling timely intervention and sustainable crop protection strategies.