← back to index
02PROJECTS

AgroSense.

team · MIT-WPU PBL4 · ay 2025–26

Aerial
UAV photogrammetry · orthomosaic · ExG / VARI
Ground
ESP32 · Modbus soil sensor · GPS waypoints
Inference
on-device ML for NPK estimation
Dashboard
Firebase · Vercel · multilingual AI assistant

Pipeline

Aerial layer (RGB → orthomosaic → vegetation indices → zoning)

The UAV captures RGB imagery over a target plot. The pipeline:

  1. RGB capture + photogrammetry — stitch overlapping captures into an orthomosaic.
  2. Vegetation indices — compute Excess Green (ExG) and Visible Atmospherically Resistant Index (VARI) from the RGB orthomosaic. RGB-only because dedicated multispectral sensors weren’t available.
  3. Zoning — K-means cluster the index map into management zones (low / mid / high vigour).

Ground layer (ESP32 + Modbus + GPS waypoints)

An ESP32 ground module carries a Modbus soil sensor and a GPS receiver. The rover follows a waypoint plan over the zones identified from the air. At each waypoint it samples the soil and runs an on-device ML model that estimates NPK (nitrogen / phosphorus / potassium) from the multivariate sensor reading.

Dashboard

Firebase + Vercel dashboard surfaces the orthomosaic + zoning overlay alongside the ground-truth NPK readings. A multilingual AI assistant answers farmer queries in their preferred language and grounds answers in the current plot data.

Review paper

Alongside the build, I authored a review paper on affordable, industry-standard precision-agriculture adoption for smallholder farms in Maharashtra — surveying what’s realistically deployable at that scale and cost.

Team

Heramb · Pranav · Paresh · me — MIT-WPU Department of Electronics & Communication Engineering, AY 2025–26.