Early detection of Chickpea Ascochyta Blight using Hyperspectral imaging Coupled with Machine learning

Fungal diseases such as Ascochyta pose major threats to chickpea production, causing significant losses if not detected early. Conventional diagnostic methods, including visual inspection and molecular assays, are often time-consuming, subjective, and ineffective for early detection of infection. This study investigates the use of hyperspectral imaging (HSI) combined with machine learning for early, non-destructive detection of Ascochyta blight in chickpea leaves, an application that remains underexplored in previous research. Hyperspectral data in the 400–1000 nm range were acquired under controlled laboratory conditions from artificially infected chickpea plants. In this study, we developed a new comprehensive processing pipeline to address critical challenges associated with hyperspectral data, including noise, artifacts, and illumination variations. Subsequently, unsupervised learning approaches, such as K-means clustering, were employed to construct a clean, well-labeled database of mean leaf spectra. Using this refined dataset, we evaluated a classification framework based on supervised learning models, leveraging selected vegetation indices, visible and infrared spectral bands, along with features derived from statistical analyses. The proposed approach achieved an overall classification accuracy exceeding 95% in distinguishing healthy chickpea plants from those infected with Ascochyta blight. Results demonstrate that HSI can capture subtle physiological changes in leaves before visible symptoms appear, offering a reliable and scalable tool for precision agriculture. This study contributes a promising step toward AI-powered early disease detection in chickpea farming, enabling timely interventions, reducing fungicide use, and supporting sustainable crop protection strategies. Future work will focus on real-world deployment and cost-effective integration into existing monitoring systems.