UAV Target Detection Based on Improved YOLOv5 Algorithm

UAV target detection has been the focus of attention and difficulty in the field of monitoring and warning. Currently, the mainstream target detection algorithm (TDA) based on neural networks are mainly designed with the application to VOC datasets in mind, and the detection accuracy in specific scenarios falls far short of the requirements. Aiming at the specific detection scenarios of UAV target detection in complex background, a UAV TDA based on improved YOLOv5 is proposed. In order to solve the weak clustering problem in the anchor frame screening process, K-means++ clustering algorithm is used to reselect the anchor frame clustering method. Focal-EIOU Loss is used to accelerate the convergence speed of the model; and the Soft-NMS algorithm is used to solve the misdetection problem occurring in the bounding box screening. The ECA attention mechanism is added to the backbone network to avoid dimensionality degradation of the model in the learning process; in the neck network, the BiFPN structure is used to realize more efficient multi-scale fusion. Improved YOLOv5 TDA is trained and tested on the dataset, and the results show that: the accuracy is as high as 97.8%, the regression rate is 97.9%, the mAP@.5 reaches 98.1%. Compared with the original YOLOv5 TDA, the present method improves the accuracy by 6.3%, the regression rate by 6.2%, the mAP@.5 by 6.0%, and mAP@.5:.95 by 4.4%. The improved YOLOv5 TDA performs at a high level in all model evaluation metrics with excellent TDA performance.