Examining the Research Gap in AI-Powered Agricultural Disease Detection for Nigerian Crops

Plant disease causes an estimated 30-40 percent annual crop loss in Nigeria, and the absence of accessible AI-based disease detection tools adapted to locally prevalent crop diseases and imaging conditions represents a critical research gap. This study develops and evaluates a convolutional neural network-based disease detection system for three economically important Nigerian crops: cassava, yam, and maize, and analyses the performance gap relative to existing international crop disease detection research. A dataset of 22,600 crop leaf images was assembled from field data collected across Enugu, Ondo, and Benue States, covering 14 disease classes and healthy plant conditions. Images were captured under varying Nigerian field lighting conditions. A ResNet-50 model fine-tuned on ImageNet pre-trained weights was trained and compared against MobileNetV3 (for edge deployment) and a custom CNN. Extensive data augmentation addressed class imbalance. ResNet-50 achieved 93.4 percent top-1 accuracy and 97.1 percent top-3 accuracy on the held-out test set. MobileNetV3 achieved 90.1 percent accuracy with 80 percent fewer model parameters, enabling deployment on entry-level Android smartphones. The study identifies a systematic performance gap attributable to non-standardised field imaging conditions and recommends a standardised image capture protocol for Nigerian field agricultural AI research. Deployment of the MobileNetV3 model as a smartphone advisory app for smallholder farmers in the cassava belt is proposed as a priority application.

Keywords: plant disease detection, convolutional neural network, cassava, Nigeria, agricultural AI