Head of the Department of Computer Science at Altai State University, Denis Kozlov, and student from the Institute of Mathematics and Information Technology, Zlata Tikhomirova, discussed a project focused on creating a vegetation map and counting sunflowers in agricultural fields.
In 2024, Natalya Ovcharova, an associate professor from the Department of Botany at Altai State University set us a task of developing a method to identify and count cultivated plants by a drone using sunflowers as an example. She provided the necessary data, and we began to work. A team of five students was formed to work on this project, including Zlata Tikhomirova, Sergey Menshikov, Yaroslav Yazykov, Gennady Kazakov, and Alexey Kovalenko.
To implement the project, it was decided to segment the images into two classes: useful vegetation and background, with everything else in the photo assigned to the latter. This task could be addressed by "additional training" of the U-net segmentation neural network, but first, the students had to painstakingly annotate the images, creating a mask that separated the sunflower from the background.
"In addition, we managed to implement a programmatic count of sunflower specimens in a photo. After that, the trained neural network was 'wrapped' in a web service capable of accepting original photos and issuing a vegetation map and the number of sunflowers," said Zlata Tikhomirova.