UAV Weed Detection Function in a Smartphone-Based Flight Computer

Effective weed control is essential in crop production. To achieve this, weeds needed to be identified correctly. Using UAV coupled with deep learning algorithms like YOLOv5 allows farmers to survey paddy fields with minimal toil. A weed detection feature was added to a smartphone-based flight computer that droves a custom-built UAV. The idea is for farmers to look at images with detection results on the phone when the drone returns. The weed detection uses a YOLOv5s model, which was validated and showed an overall P, R, and mAP50 of 94.4%, 84.5%, and 89.3%, respectively. The system was tested by sending the drone into autonomous waypoint flight multiple times. It is observed that the weed detection algorithm could detect and identify weeds when the UAV is at lower altitudes or is hovering and moving slowly.