→ next · ← back · F full screen · T timer · N notes
0:001 / 12
INNOCHAMP Season 2 · Grand Finale · Team IC2-CAT1-049

Wise Agri Growth

Find the sick tree. Treat only the sick tree.
The F450 drone
The drone camera's view of the leaf mat
The treatment rover
Sujin S · Class 8 · St. Antony's Public School, Tirunelveli
The problem

100 trees. 1 is sick.
All 100 get sprayed.

99 trees of wasted chemical and money
Poison soaks into the soil
The person spraying breathes it in
Why? Farmers can't see which tree is sick.
My solution

Find the sick tree. Treat only that tree.

Both drones
Drone · the eyes
Scouts the trees with its camera
→
The AI marking sick leaves
AI · the brain
Spots the sick leaves
→
The rover
Rover · the hands
Treats only that tree
The brain

Tested on my own garden.

Drone camera view of the 3 by 3 leaf mat
My drone's camera over the leaf mat (9 Oct).
89 %
right on 44 leaves from my own garden
8 / 9
leaves right from my drone's camera
About 10 AI models tested on my leaves; I kept the best: SigLIP (Google, 400 million parameters)
Finding the same tree again

No GPS. The drone makes its own map.

Every tree found from above and numbered by row and tree
1Fly highFind every tree, number it by row
4,5 the drone's photo
2Star trackerMatch the pattern of trees and gaps → row 4, tree 590–100 % the right tree, in simulation
click to watch big
3Count trunksRover drives the row with its ultrasonic sonar to tree 5
Live on this table

This mat is a small orchard.

1
2
3
4
5
6
7
8
9
1 · Click the 4 cornersEach square becomes a tree
2 · The AI checks every squareSick squares turn red
3 · Tap a red squareThe rover turns to face it
The hands

Medicine at the roots. Not in the air.

3 sSoftenWater loosens the soil
2 sNeedleLowered slowly
3 sDoseMedicine at the roots
≈10 sDoneNeedle up, safe
✗ Sprayingdrifts into the air: leaves, soil, people✓ Rover: root dosemedicine goes into the roots only
Fail-safe: STOP or lost signal → it stops, needle up
Inside the rover, every part labelled
Inside my rover, every part labelled: ESP32 brain, 3 pumps, 2 servos, motor drivers, relays, both battery packs.
Impact · precision agriculture

1 sick tree in 100 → 1 treated, not 100.

Spray every tree: 100 trees of chemical
Wise Agri Growth: 1 tree, at the roots
89 %right on my garden's leaves
8 / 9right from my drone's camera
≈10 sto treat one tree
Less cost · less poison in the soil · caught early, before it spreads
Scale

The software can guide the spraying drones farmers already use.

Not only mango:
MangoApplePomegranateBananaGrapeGuava
Next: my bigger F450 drone (built, flying since 24 Oct), then tests in real orchards with farmers.
The F450 drone among model trees
100 trees. 1 sick. 1 treated.

AI eyes in the sky.
Precise care on the ground.

Sujin S · St. Antony's Public School, Tirunelveli · Wise Agri Growth
The rover, the F450 among model trees, and the XIAO drone
For questions · Accuracy

Measured, not assumed.

Photos ofImagesRight
Leaves on the living plant (field)26593 %
Web photos (PlantDoc)60094 %
Lab photos (PlantVillage)55593 %
Picked leaves on paper28890 %
My garden's mango leaves4489 %
Most mistakes are missed diseases, not false alarms: 31 false alarms in about 900 healthy leaves.
Weak so far: citrus 21 % sugarcane 69 %. They need their own tuning.
For questions · The drones

Two drones I built.

Small drone (XIAO ESP32-S3)
69 g · my own flight firmware · live camera over its own WiFi · my first drone
F450
0.9 kg · Radiolink F722 with Betaflight · self-levelling · FS-i6 radio with tested failsafe · its own camera at 27 frames/s · first flight 24 Oct
I practise in a flight simulator I built from its own flight logs.
The XIAO drone and the F450
click to watch big
For questions · How I built it

From one small drone to three machines.

JuneBought a DM002HW and tried to crack its code with Ghidra. It failed, so I started my own
1 AugFirst drone firmware: my small XIAO drone
8 SepWise Agri Growth v5: the first save
28 SepRover drives, sprays and injects from a phone
1-3 OctAI (SigLIP) tested on my garden; the no-GPS orchard map
10 OctRound 1
16 OctSelected for the Grand Finale
24 OctF450 first flight
25-27 OctRover rebuilt and rewired
29 OctPainted, labelled, turning fixed, counts trunks
31 OctGrand Finale
337 saves on 37 different days since 8 Sep
107 Sep
114 Sep
621 Sep
9328 Sep
675 Oct
6612 Oct
3919 Oct
5526 Oct
Setbacks I fixed XIAO's camera and antenna died: the F450 got its own camera A capacitor shorted and the F450's 5 V converter smoked: rescued with a new 5 V supply Rover failed: rebuilt on a new chassis and rewired AI: models trained on photos of leaves on paper called every real tree sick; I tested about ten before SigLIP: 89 % on my garden
Development cost so far: about ₹50,000
Idea, design, hardware and every test: mine. The code is written with an AI coding assistant, directed and tested by me.
Wise Agri Growth

Thank you.

Find the sick tree. Treat only the sick tree.
Sujin S · Class 8 · St. Antony's Public School, Tirunelveli · Team IC2-CAT1-049