What is it about?

Machine learning is a way to program a computer to do something without explicitly telling the computer what to do. Instead, it learns from examples. Our goal is to create public data for use in developing machine learning algorithms to solve problems in the field of agriculture. In this work, the examples are digital images of plants with a corresponding label that identifies the specific variety of plant in the image, and it takes a lot of images - tens or hundreds of thousands per variety - to develop a good algorithm. Finding or creating such data for agricultural applications is very time-consuming and thus costly. As a result, we created a robot that can image crops and weeds from nearly every angle and several plants at the same time. The resulting images are automatically labelled and uploaded into our database and made available for machine learning developers and researchers to use.

Featured Image

Why is it important?

The growing global population requires more efficient growing of food, while we need to protect our environment. To increase yields from fields a digitalization of agriculture is anticipated, implementing new technologies such as unmanned vehicles and machine learning into farmer's workflows. These approaches require a large volume of diverse data to be successfull. In this work we present a system that can significantly contribute to creating the needed datasets, as it offers flexibility in what and how plants are being imaged, while providing a high degree of automation and image collection rate.


Building the EAGL-I system and publishing about it, really laid down the foundation of our digital agriculture project "TerraByte". It is a very potent system in itself and the data it produced for us, has helped us and other scientists in answering research questions in agriculture. Moreso, it has inspired us to create more data collection systems and datasets. We cannot wait to see what our upgraded version of EAGL-I will do for us.

Michael Beck
University of Winnipeg

Read the Original

This page is a summary of: An embedded system for the automated generation of labeled plant images to enable machine learning applications in agriculture, PLoS ONE, December 2020, PLOS,
DOI: 10.1371/journal.pone.0243923.
You can read the full text:

Open access logo



The following have contributed to this page