Student
Stu
2
AD
Aarav Deshmukh
Student
Proposal submission
SIH1601 — AI-based Crop Disease Detection for Small Farmers
Demo data
Under AI Analysis
Proposal title
Technology stack
Proposed solution
54 words
A mobile-first Android application that lets a farmer photograph an affected leaf and receive an on-device disease diagnosis within two seconds, along with severity grading and locally available remedies priced in INR. The model runs fully offline via a quantised TFLite model, and syncs anonymised diagnoses to a central dashboard when connectivity is available.
Innovation
36 words
On-device quantised vision transformer tuned for Indian crop varieties, combined with a Marathi/Hindi voice output layer so that low-literacy farmers can use the app without reading. A crowd-sourced outbreak heat-map warns neighbouring villages of spreading infections.
Technical approach
37 words
EfficientNet-B0 backbone fine-tuned on PlantVillage plus a locally collected dataset of 14,000 images from Pune district. Quantisation aware training reduces model size to 6.8 MB. Flutter front end, FastAPI sync service, PostgreSQL + PostGIS for outbreak clustering.
Target users
10 words
Small and marginal farmers, Krishi Vigyan Kendra officers, agri-input retailers
Expected impact
20 words
Reduces crop loss from delayed diagnosis by an estimated 18-25% and cuts unnecessary pesticide expenditure for a typical two-acre holding.
Scalability
26 words
Model registry allows adding new crops without releasing a new app version. Sync service is stateless and horizontally scalable; outbreak clustering runs as a scheduled job.
Implementation approach
25 words
Week 1-2 dataset and model, Week 3 mobile app, Week 4 sync backend and dashboard, Week 5 field pilot with 30 farmers in Tathawade region.
AI Assistant