AI System Counts Sunflower Seeds in Seconds
Russian researchers have developed a computer vision system that automatically counts and evaluates the quality of sunflower seeds on harvested flower heads, dramatically accelerating crop breeding and yield analysis.
A single sunflower head can contain anywhere from several hundred to more than a thousand seeds, and manually separating filled seeds from empty ones typically requires about an hour of painstaking work.
Researchers from the All-Russian Research Institute of Agricultural Biotechnology and the Moscow Center for Advanced Technologies trained the neural network on a dataset of more than 1,000 images containing approximately 260,000 sunflower seeds. The system achieves an identification accuracy of 88%.
The researchers plan to adapt the model for field use, including deployment with drones. The system is already available through a public Telegram bot, while the annotated dataset and trained model have been released as open-source resources.
