Prototype showcase purpose
Recently, our team in Plant.id started working with datasets of aerial photographs of agriculture fields. The purpose of this new machine learning model is to locate and distinguish different plant species in the field covered with crops and various species of weeds. The aim is to help applying herbicide, pesticide or fertilizer more effectively.
This UAV photo shows a cropped detail of a much larger image of a field with the area of a few hectares. Although the photo has a rather low resolution, the model can identify areas with seedlings of sunflowers (Helianthus annuus) and various weedy species.
To train the machine learning algorithm, we let humans to draw areas with the location of particular object(s) (red areas). The blue areas represent the resulting locations where a particular weed or crop occurs (actually, it also distinguishes bare soil class). The probability of the occurrence is marked by the intensity of the blue color).
We are ready to train the machine learning model to identify various other plant species or different objects.
The blue areas can be returned in the CSV, JSON or another format suitable for agriculture sprayer. We assume that the client will set up the “dosage” level based on the different combinations of weeds and crops in a given square.
Proposed application and business model
There are three possible directions for further development
- A web app (like Plant.id) for a manual upload of the UAV photos
- An API similar to our plant identification API for close-up photos
- A PC app which you can run directly in the field
We propose two business models
- pay per processed pixel
- annual license and paid upgrades for different species
Let us know which technical direction or business model do you prefer: firstname.lastname@example.org