Scope:
2,000 images, 60,000+ polygons
Project:
To to develop a training dataset for a ML model capable of distinguishing weeds from agricultural species and identifying growth point(s) according to the type of weed. The model can be integrated in a laser weeds extermination device.
Challenge:
Industry expertise required in order to classify weeds and agricultural species, identify types of weeds (monocot, ) and define growth point(s).
Solution:
Due to the complexity of the task, we limited the number of workers to those with specific knowledge and skill, whose job was checked by validators – experts in the field.
Annotating Weeds | Case Study
The project was to develop a training dataset for an ML model capable of distinguishing weeds from agricultural species and identifying growth point(s) according to the type of weed. The model can be integrated into a laser weed extermination device.

What problem are you trying to solve?

We're tackling an ageless agricultural problem: weed management. Weeds have always caused great yield losses in crop fields and herbicides have been the first solution most farmers resort to for a while now. At LUXEED, we're developing an autonomous weeding robot that eradicates the weeds in certain crop fields using a localized thermal method that has minimal to no environmental impact.

What is your solution? What is unique about it?

We're burning the weeds using a laser beam to burn specific parts of the plants and kill them. Our technology relies on artificial intelligence and machine learning to be able to do so. This type of technology is relatively new worldwide with one entry to the market in the whole world till now.

How are you going to use the labeled data to help you solve the problem?

The labeled data at hand will allow the robot to correctly recognize the crop from other plants, as well as locate desired areas within the weeds.

What is the main challenge in collecting and labeling the right dataset and how are you solving it?

Collecting a dataset is a time-consuming and repetitive task that has to be done under a lot of different conditions in order to have a robust model. Labeling demands even more effort since it is an iterative process that requires a certain level of expertise and an eye for detail.