Recognition of Soybean Crops and Weeds with YOLO v4 and UAV
摘要
This paper describes an approach to the detection and recognition of crops and weeds in an agricultural field using data obtained from unmanned aerial vehicles (UAVs) and the YOLO v4 neural network. The advantage of using YOLO v4 is the recognition of objects and the names of each object. The creation of a data set for training a neural network is described, its preprocessing, training of a neural network, and the results of recognition of crops and weeds in a soybean field are described. Neural network quality for object confidence threshold ≥0.25: precision = 0.81, recall = 0.82, F1-score = 0.82, average IoU = 58.40%, mAP = 31.81%. Also in the corresponding section, the proposed approach is discussed in comparison with the approach that does not require manual data annotation.