Rice Weed detection method on YOLOv4 convolutional neural network

Aiming at the precise spraying of chemical pesticides on related weeds in precision agriculture, this paper proposes a method to detect 12 kinds of rice weeds by using YOLOv4convolutional neural network. By taking pictures of the natural environment growth in the field, labelImg manual data annotation, and deep learning framework training of Darknet, it has a good effect on weed detection and recognition. The experimental results show that the detection accuracy of YOLOv4 model is 97%, the recall rate is 81%, the F1 value is 0.89, the average detection time is 377ms, and the cross merge ratio is 81.86%. Under the same conditions, the mAP value index is 11.6% higher than that of YOLOv3, and the detection time is increased by 11ms. This detection accuracy and detection time meet the needs of precision agriculture, and provide theoretical and technical support for the rapid identification of rice weeds and precise spraying of pesticides by machinery.