Scaling AI4NYP+: Edge-AI and SMS-Powered Crop Disease Detection for Smallholder Farmers in Enugu State, Nigeria

Smallholder farmers across Southeastern Nigeria cultivate pepper, tomato and maize as staple crops that supply household food, local markets and seasonal incomes. These families, often farming on plots of less than two hectares operate in fragile systems where a single disease outbreak can wipe out a season?s income and push households into food insecurity. In Enugu State, limited public extension capacity, poor rural connectivity, and low digital literacy constrain farmers? ability to identify crop diseases early and act appropriately. As a result, disease symptoms are frequently recognized only at advanced stages, responses are delayed or inappropriate, and pesticide misuse is common: farmers spray excessively or apply inappropriate chemicals because they lack timely, trusted guidance. The consequence is predictable, avoidable yield losses, higher input costs, health risks for farm families, and environmental harm from chemical runoff.
Existing agricultural advisory services are sparse, overstretched, and often urban-centric. Many digital agritech solutions assume internet access and smartphone ownership; in rural Enugu large numbers of smallholders lack either. Cloud-dependent AI systems, smartphone apps, or web-based decision-support tools therefore exclude the most vulnerable users which includes women, older farmers, and low-literacy households, who make decisions with limited, localized information. There is a clear demand for an approach that is low-cost, offline-capable, local language-centered, and integrated into everyday farming workflows.
AI4NYP+ was developed precisely to bridge these gaps. It is an offline-first, solar-powered Edge-AI system designed for continuous in-field monitoring and instant, low-bandwidth advisory delivery. The system pairs rugged, low-cost compute (Raspberry Pi-class devices) with dual cameras that capture leaf images throughout the day. A compact, optimized computer-vision model (YOLOv11n-based) runs on-device to detect early symptoms across seven classes (leaf spot, powdery mildew, yellowing, curling, blight, mosaic virus, healthy). When the model detects a meaningful anomaly above conservative confidence thresholds, an SMS advisory, written in Igbo with an English fallback, is dispatched to farmers in the affected plot. The SMS is short, actionable and grounded in local extension best-practices; planned upgrades add two-way SMS so farmers can acknowledge alerts, request clarification, or flag uncertain detections.
AI4NYP+ has moved beyond lab proofs into real-life deployment. The system was trained initially on an 8,000+ image dataset focused on Nsukka Yellow Pepper and deployed in three communities in Nsukka LGA (Eziani, Nguru, Agbani), serving 120+ smallholders. Early impact evidence is promising: participating farmers reported 17?25% increases in yield and approximately 30% reductions in pesticide misuse. These results demonstrate both the technical viability of edge inference in constrained environments and real-world acceptance among farmers who previously had little access to reliable diagnostics.
Despite this success, the current deployment covers limited geography and a single crop. Scaling is necessary to expand the social and economic benefits to more farmers and to validate the approach across crop types and agroecological conditions. Extension to tomato and maize requires collecting and annotating diverse field images (we plan 25,000+ additional labeled images), retraining multi-crop models optimized for edge deployment, and enhancing the SMS advisory system with crop- and disease-specific guidance. Scaling also involves operational investments?bulk procurement of cost-effective deployment kits, community champions to manage devices, and SMS gateway credits to sustain two-way communication.
The significance of scaling AI4NYP+ is multi-dimensional. First, on livelihoods and food security: earlier and accurate detection reduces time-to-treatment, prevents yield collapse, and increases household income stability. Second, on public health and environment: reducing indiscriminate pesticide use protects farmer health, lowers residue in food, and reduces contamination of soil and water bodies?contributions to local environmental sustainability and climate resilience. Third, on inclusion and gender equity: the design intentionally targets low-literacy and non-smartphone users (many of whom are women), delivering information in Igbo and working through cooperatives and local extension networks to ensure equitable access. Finally, on capacity-building: the project trains local technicians, extension officers and community champions, strengthening the local ecosystem and increasing the probability that the intervention will persist after project funding ends.
Gaps that scaling must address are both technical and institutional: increasing model robustness across lighting, varietal and agronomic differences; ensuring data governance and farmer consent; preventing reliance on a single technology vendor; and embedding the solution within existing agricultural advisory systems so that alerts translate into practical, affordable on-farm actions. Our scaling plan addresses these: we will implement active learning to prioritize uncertain field examples for manual labeling; apply privacy-preserving data handling and informed consent protocols; recruit and train local champions and extension officers for maintenance and user support; and co-design SMS advisories with farmers and extension agents to ensure recommended actions are feasible and affordable.

Smallholder farmers in Enugu State face significant crop losses from diseases that go unrecognized until advanced stages. Limited extension services, poor internet connectivity, and low smartphone ownership prevent timely diagnosis and intervention. As a result, farmers either delay action or over-apply pesticides, with negative effects on yields, household income, health, and the environment. AI4NYP+ addresses this gap by offering an offline, solar-powered field monitoring device that automatically detects disease early and sends clear SMS instructions in Igbo and English. This reduces time-to-action, prevents yield loss, and curbs pesticide misuse. Scaling the system to additional LGAs and crops (tomato and maize) will expand access to life-saving agricultural intelligence for underserved communities, improving food security and reducing environmental harm.

Primary objective: Scale AI4NYP+ to reach 420+ smallholder farmers in Enugu State within 12 months by expanding geographic coverage and crop detection capabilities while maintaining responsible and inclusive design. Sub-objectives:
1. Contribute to local African Datasets by collecting and annotating 25,000+ high-quality images for tomato, maize, and, in addition to the existing pepper datasets, across diverse farms (Months 1?6).
2. Retrain and optimize multiclass YOLO models for Raspberry Pi devices and deploy them to new sites (Months 4?8).
3. Upgrade the SMS advisory system for two-way interaction, localized language support (Igbo + English), and context-sensitive recommendations (Months 6?10).
4. Onboard and train 300+ additional farmers via cooperatives and extension officers, with emphasis on women and vulnerable groups (Months 8?10).
5. Monitor impact (yield, pesticide use) and iterate models and advisories based on farmer feedback (Months 10?12).