Open-Earth-Monitor Global Workshop 2026

Fungal diseases such as Ascochyta remain a major obstacle to chickpea production, leading to significant yield losses if not detected early. Whilst hyperspectral imaging (HSI), combined with machine learning, has demonstrated strong potential for early and non-destructive detection under controlled laboratory conditions, its transferability to real-world field environments remains a major challenge. This study aims to validate detection models developed in the laboratory under field conditions using multispectral images acquired by a drone. Following promising results obtained using hyperspectral data (400–1000 nm) and advanced machine learning pipelines, we have extended our approach to drone-based multispectral detection to assess its robustness in real-world agricultural scenarios. Field data were acquired using drone-mounted sensors capturing key spectral bands relevant to vegetation health and stress detection. A comprehensive processing pipeline was implemented, comprising radiometric correction, image pre-processing, vegetation index calculation, and feature extraction. The previously developed classification framework was adapted and applied to multispectral data, incorporating both spectral and statistical features. The results demonstrate that the proposed approach can be successfully applied from the laboratory to field conditions, achieving high detection performance with a classification accuracy of over 90% in distinguishing healthy chickpea plants from infected ones. Furthermore, the system proved capable of detecting signs of infection at an early stage in the canopy, despite environmental variability such as changes in light intensity and ground background effects. These results confirm the feasibility of deploying AI-based disease detection systems using drone-based multispectral imaging in real agricultural environments. This work represents a significant step towards operational precision agriculture solutions, enabling large-scale monitoring, early intervention, reduced chemical inputs, and improved crop management strategies. Future work will focus on validation over multiple seasons, integration with close-range detection, and extension to other plant-pathogen systems.