A neural network that determines the weight of the salad from the photo

ABOUT
ABOUT
ABOUT
ABOUT
ABOUT
ABOUT

Customer

01/
iFarm is a Russian IT startup that builds vertical software-controlled farms for growing berries, vegetables and herbs.
02/

A task

Нейросеть
To teach a neural network to distinguish between Romano salad in a photo and predict its weight.
This will help the company monitor plantings around the clock and know when the plant is stunted. This means:
- Notice when lettuce starts to lag behind
- Automatically regulate the microclimate in the greenhouse
03/

What have we done

Interesting fact

Determining the weight of the salad, the neural network calculates the area of the green area in the image.

How it works

04/
Stationary cameras take pictures of salad on the shelves
Images are sent to the AI system EORA Vision Hub
The neural network determines the parameters of the plant and compares them with the forecast
If the indicators are very different, the neural network tells about it
IFarm system regulates the microclimate
in the greenhouse

How it works

04/
Stationary cameras take pictures of salad on the shelves
Images are sent to the AI system EORA Vision Hub
The neural network determines the parameters of the plant and compares them with the forecast
If the indicators are very different, the neural network tells about it
IFarm system regulates the microclimate
in the greenhouse
1
2
3
4
5

Our model vs reality

05/
Predicted weight
Maximum weight that day
Average weight that day
Minimum weight on this day
Weight in grams
Growth day

The problems we faced

05/
Data markup
Decision
Time spent
We taught the neural network to see the outlines of the bushes. But the lettuce grew and they merged
We began to enclose the bushes
in rectangles. Marking sped up to 2 minutes.
It took up to 15 minutes to mark the contours in one photo.

What is the value of the project

07/
Ecology
Saving
Plants need fewer pesticides and fertilizers.
The cost of the salad is lower: the care is partially automated.

Project team

08/
Techlead
Konstantin Kubrak
Computer Vision Engineer
Aelita Shaikhutdinova
Project manager
Nadezhda Zagvozkina
Backend developer
Andrey Chertkov

Our plans

06/
Automate the collection of plant data
Start processing photos on the spot, without sending them to the server
Use a drone with a camera to shoot plants.
Contact us