AI-Based Cassava Leaf Disease Detection to Support Smallholder Farmers in Ghana

Late detection of cassava leaf disease is a major cause of crop loss among smallholder farmers in Ghana, where manual field inspection is often the only diagnostic tool available. This paper presents a lightweight, mobile-deployable image classification system for early cassava disease awareness. We apply transfer learning with a MobileNetV2 convolutional neural network, pretrained on ImageNet, to the public iCassava leaf image dataset across five classes: healthy, Cassava Bacterial Blight (CBB), Cassava Brown Streak Disease (CBSD), Cassava Green Mite (CGM), and Cassava Mosaic Disease (CMD). The model is paired with a farmer-facing advisory layer that translates each prediction into short, actionable guidance. On a held-out test set (n = 1,885) the model achieves 76.71% overall accuracy (test loss: 0.6479; weighted F1: 0.75; macro F1: 0.64) after 5 training epochs with the MobileNetV2 base frozen. Per-class analysis reveals substantial variation in performance: the majority class (CMD) reaches 0.90 recall, while the minority class (CBB) reaches only 0.30 recall, driven largely by confusion with CBSD. We situate these results against prior work on the same dataset lineage, discuss the model's limitations — including its reliance on secondary, non-Ghana-collected imagery and its uncorrected class imbalance — and outline a path toward field validation and offline mobile deployment via TensorFlow Lite. This work demonstrates that existing, publicly documented deep learning techniques can be adapted at low cost into a locally-relevant decision-support tool for Ghanaian agricultural extension.   Keywords: cassava disease detection, transfer learning, MobileNetV2, precision agriculture, Ghana, smallholder farming, class imbalance