Document current crop-disease, water-management and irrigation practices among smallholders.
Research foundation
From a local research question to a field-ready platform.
AgriSentry is informed by Taiwo Oluwadamilola Olaniyi's March 2025 MSc proposal on machine learning, public weather data and precision irrigation for smallholder maize and cassava farmers in Ogbomoso, Oyo State.
Evidence status
Proposal—not study results
The six-month farmer trial and local model evaluation are proposed. This site does not claim completed trials, measured water savings, yield gains or validated detection accuracy.
Research problem
Two costly decisions, one accessible interface.
The proposal focuses on late visual detection of crop disease and inefficient schedule-based irrigation. Its design constraint is practical: work through an ordinary smartphone browser, keep the farmer workflow simple, and use SMS where data access is limited.
Design a browser-based experience that accepts camera photos without requiring an app installation.
Implement server-side YOLO-Nano screening for maize and cassava disease symptoms.
Evaluate weather-led automatic irrigation against usual farmer practice using water use and yield.
System translation
What the research means inside the product
Status shown for each path
Interactive demo • validation pending
Crop condition workflow
- 1Farmer captures one affected leaf
- 2Image is processed by server-side YOLO-Nano
- 3Result links to symptoms, action and local support
The proposal calls for PlantVillage pretraining, local Ogbomoso fine-tuning, agronomist labels and a held-out local test set. The current browser demo does not run that trained model.
Planning tool • automation proposed
Precision irrigation workflow
- 1Read recent rain and a three-day forecast
- 2Update the seven-day root-zone water ledger
- 3Trigger a relay-controlled pump and send an SMS
The site exposes the decision math. Live weather ingestion, calibrated crop thresholds, a farm pump relay and automatic SMS delivery require a production deployment and field validation.
Water Balance Theory
Account for every millimetre
Δ soil water = rainfall + irrigation − evapotranspiration − drainage
The proposal estimates root-zone moisture from recent rain, evapotranspiration and prior irrigation when a soil sensor is unavailable. AgriSentry's irrigation page turns this principle into a transparent planning calculation.
Technology Acceptance Model
Useful and easy enough to adopt
The interface prioritises visible outcomes and short, familiar actions: open a browser, take a photo, read a recommendation, or receive an SMS. No specialist hardware or installed mobile app is assumed for the core farmer experience.
Basic smartphoneLow bandwidthLocal languageSMS fallback
Proposed evaluation
A six-month field trial designed around local use.
The protocol proposes 60 purposively selected maize or cassava farmers: 30 using the intervention and 30 continuing usual practice, distributed equally across Ogbomoso North, South and Central.
60
farmers
3
zones
6
months
Disease screening
Precision, recall, F1 score and accuracy on a held-out local image set
Water efficiency
Treatment versus control water use, measured with flow meters and logs
Crop outcome
Harvest weight and percentage yield difference between study groups
Adoption
Usability feedback interpreted through usefulness and ease-of-use
Understandable consent
Explain participation in Yorùbá, with signature or thumbprint consent.
Voluntary and private
Allow withdrawal at any time and remove names or identifiers from publications.
Approval before research
Obtain university ethics clearance and local permission before recruiting or collecting trial data.
Training-ready data governance
Retrieve evidence now. Train only after permission and validation.
The top-tier assistant can search IITA's CKAN catalogue for research references. It does not silently train on catalogue files or farmer records. A model-training run must record provenance, licence, consent, de-identification, validation status and an accountable approver first.
AgriSentry system records
Eligible records must be purpose-limited, de-identified and explicitly approved. Account, contact and notification data are excluded by default.
IITA CKAN catalogue
Dataset metadata is searchable through the protected assistant. Each file's licence and scientific suitability must be verified separately before training.
NiMet data
The connector accepts a server-managed endpoint and token after NiMet grants API/data access. No unofficial endpoint is substituted.
Explore the implementation
See the research logic in action.
Use the transparent irrigation planner or review how production data and research participation must be protected.
Source: Olaniyi, T. O. (2025), Leveraging Machine Learning and Public Data Sources for Smart Crop Condition Monitoring and Precision Irrigation: A Software Approach for Smallholder Farmers in Ogbomoso, Oyo State, Nigeria, MSc research proposal, University of Kigali. Research details are presented as proposed methods unless stated otherwise.