Detection of Sunflower Plants in UAV Photos

The task of detecting sunflower plants in the field of UAV image to estimate an important agrotechnical parameter - the density of seedlings - is being solved in this work. The networks of YOLOv5 family were used as a detector model. The question of learning the quality in small samples is raised. The best result was obtained by the YOLOv5m network in an augmented sample of 363 photos. When the recognized images are being noised with weeds, the detection quality decreases. On the whole, the YOLOv5m model copes with the task, and further improvement of the detection quality is related to the expansion of the training data corresponding to different growth conditions of sunflower plants.