To learn more about agribusiness in Ukraine, follow us on Facebook, on our channel in Telegram, and subscribe to our newsletter.
Kernel introduces AI for crop condition assessment
13 May 2021, 08:48
The company Kernel applies artificial intelligence (AI) to assess the density and quality of crops in photos taken from a drone. The company says that the algorithms have already been tested and the full-scale implementation in Kernel agricultural production is planned for the current season.
Kernel IT Director Andriy Pishyy marks that the use of modern technologies in the company's agribusiness, including automation and integrated implementation of precision farming systems, enables Kernel to accumulate data on the fields and technical operations performed, etc.
"Satellite images, drone photos, and ground survey data are processed and stored in the company's IT infrastructure. Their application involves the development of algorithms to carry out in-depth, objective and at the same time real-time analysis of the field to solve complex agricultural production problems and make effective business decisions," he comments.
For this purpose, Kernel established the Data Science department in the IT department, Andriy Pishyy adds.
Over the year, the Data Science specialists added new software products to Kernel's #DigitalAgriBusiness information ecosystem. The first thing they did was to integrate a model for counting grains in a corn cob and seeds in a sunflower head into the agronomist's personal application. In addition, they have elaborated a model for assessing the quality of seeding with drone photos.
Kernel's Deputy Agribusiness Director for Innovation and Digital Development Evgenii Sapizhenko emphasizes that Machine Learning, Deep Neural Network and other methods have been used to develop the algorithms. The company is committed to sharing this tool with Open Agribusiness partners and distributing it on a commercial basis, he notes.
Tests in the company's clusters have shown a good result, with an average error value of 2% being recorded. According to Kernel, the model test run is already significantly more accurate than manual field counts using existing practices.
