Artificial Intelligence and Deep Learning for Real-Time Disease Detection in Nigerian Staple Crops Using Hyperspectral Imaging

This study develops and validates deep learning models for real-time detection of major diseases in Nigerian staple crops using hyperspectral imaging, contributing original methodological advances to precision agricultural engineering for disease management. Late detection of crop diseases is a major contributor to yield losses in Nigerian staple crop production, with cassava mosaic virus, maize streak virus, and rice blast among the economically most important pathogens. Hyperspectral imaging can detect disease-induced biochemical and structural changes in plant tissue before visible symptoms are apparent, enabling early intervention. This study collects hyperspectral image datasets using a field-portable hyperspectral camera across three seasons from experimental and farmer fields in Ogun, Kaduna, and Enugu States. Ground truth disease severity ratings are obtained through laboratory molecular testing and expert visual assessment. A convolutional neural network architecture is optimised for hyperspectral disease detection, incorporating spatial-spectral attention mechanisms and domain adaptation for field condition variability. Transfer learning from international crop disease datasets improves model performance under limited Nigerian training data. Comparison with existing RGB image-based detection methods quantifies the information advantage of hyperspectral data. Findings reveal that the proposed hyperspectral CNN achieves disease detection accuracy of 93.7 percent at an average of 8 days before visible symptom onset, compared to 76 percent for RGB-based detection at visible symptom stage. The system operates in real time on an edge computing device. The study recommends integration of the hyperspectral disease detection system with UAV platforms for field-scale deployment.

Keywords: deep learning, hyperspectral imaging, crop disease detection, precision agriculture, Nigeria.