The challenge
Cassava feeds millions of households across Africa, yet leaf diseases cut yields and threaten food security. Field diagnosis is slow and expensive for smallholder farmers, and cloud-only tools fail where connectivity is weak.
What we built
SmartCassava is an AI-powered mobile prototype for cassava leaf diagnosis, severity analysis, and farmer recommendation support. A MobileNetV2 model was trained with transfer learning, converted to a compact TensorFlow Lite package, and deployed for on-device inference in a Flutter app so farmers can classify leaves offline.
- Five-class leaf diagnosis: Healthy, CMD, CBSD, CBB, and CGM
- MobileNetV2 to TensorFlow Lite for low-cost phones (~9MB)
- Rule-based severity (mild, moderate, severe) from confidence
- Local JSON recommendations and Flutter field prototype
Outcome
Farmers get a portable decision-support tool: snap or pick a leaf photo, see the disease class and severity, and receive plain-language next steps without needing a server. The work was published in Discover Artificial Intelligence (2025).