The drone detection based on improved YOLOv5

Abstract

The wide application of drones not only brings convenience to production and life, but also poses a threat to public safety. Therefore, the detection of s is crucial. However, tiny drones make it difficult to cope with traditional detection methods such as radar and photoelectricity because of their tiny size. Therefore, this paper proposed a tiny drones detection method based on YOLOv5 framework. By optimizing the size of Anchor box, embedding the Convolutional Block Attention Module (CBAM) and optimized loss function (CIoU), the detection performance of the original algorithm for drones under complex background is improved. The improved YOLOv5 algorithm is trained and tested on the self-built dataset, and its mean Average Precision, Accuracy and Recall reach 96.9%, 97.8% and 95.6% respectively. Finally, the improved YOLOv5 is used for drone detection in complex background environments. Compared with the original algorithm, it can correctly identify drone targets in harsh environments.