A novel small object detection algorithm for UAVs based on YOLOv5
Jianzhuang Li,Yuechong Zhang,Haiying Liu,Junmei Guo,Lida Liu,Jason Gu,Lixia Deng,Shuang Li
DOI: https://doi.org/10.1088/1402-4896/ad2147
2024-01-23
Physica Scripta
Abstract:Due to the advances in deep learning, artificial intelligence is widely utilized. Technologies frontier, including computer vision, represented by object detection, have endowed unmanned aerial vehicles (UAVs) with autonomous perception, analysis, and decision-making capabilities. UAVs extensively utilized in numerous fields including photography, industry and agriculture, surveillance, disaster relief, and play an important role in real life. However, current object detection algorithms encounter challenges when it comes to detecting small objects in images captured by UAVs. The small size of the objects, high density, low resolution, and few features make it difficult for the algorithms to achieve high detection accuracy and are prone to miss and false detections especially when detecting small objects. For the case of enhancing the performance of UAV detection on small objects, a novel small object detection algorithm for UAVs adaptation based on YOLOv5s (UA-YOLOv5s) was proposed. 1) To achieve effective small-sized objects detection and better detection performance, a more accurate small object detection (MASOD) structure was adopted. 2) To boost the detection accuracy and generalization ability of the model, a multi-scale feature fusion (MSF) approach was introduced, which fused the feature information of the shallow layers of the backbone and the neck. 3) Towards enhancing the model stability properties and feature extraction capability, a more efficient and stable convolution residual Squeeze-and-Excitation (CRS)module was introduced. Compared with the YOLOv5s, mAP@0.5 was achieved an impressive improvement of 7.2%. Compared with the YOLOv5l, mAP@0.5 increased by 1.0%, and GFLOPs decreased by 69.1%. Compared to the YOLOv3, mAP@0.5 decreased by 0.2% and GFLOPs by 78.5%. The study's findings demonstrate that the proposed UA-YOLOv5s significantly enhance the object detection performance of UAVs campared to YOLOv5s.
physics, multidisciplinary