Abstract: | Strawberry object detection is of great significance for intelligent monitoring and automated picking of strawberries. In this paper, a strawberry target detection method based on YOLOv4 is proposed. For the strawberry dataset collected in the complex environment, the LabelImg data type is first labeled, then the improved Kmeans clustering algorithm is used to calculate the size of the transcendental box, and finally the YOLOv4 model is trained and evaluated by the phased training method. The experimental results show that the average accuracy mAP of the test set of this method reaches 97.05%, the recall rate of ripe strawberries is Call=90.48%, the accuracy precision is Precision=98.70%, and the mean accuracy AP is 99.63%, the recall rate of unripe strawberries is Call=80.40%, the accuracy Precision=94.67%, the accuracy mean mAP is 94.47%, and the detection time of a single image is 74ms. It can meet the needs of high-precision real-time detection of strawberries. |