Small Object Detection in Aerial Images
- Resource Type
- Conference
- Authors
- Zhang, Ruoyu; Jing, Minge; Fan, Yibo; Zeng, Xiaoyang
- Source
- 2021 IEEE 14th International Conference on ASIC (ASICON) ASIC (ASICON), 2021 IEEE 14th International Conference on. :1-4 Oct, 2021
- Subject
- Components, Circuits, Devices and Systems
Computer vision
Conferences
Computational modeling
Object detection
Neck
Standards
- Language
- ISSN
- 2162-755X
Object detection is one of the main research directions in the field of computer vision. For aerial images which have many small and dense objects, more sophisticated object detection models are required to obtain satisfactory result. In this paper, S_ YOLOv5 is proposed with a more appropriate detection layer of small-scale objects. Furthermore, batch mode weighted-cluster-DIoU-NMS is adopted instead of the standard NMS to efficiently eliminate redundant bounding boxes. The experimental results show that S_YOLOv5m with improved NMS achieved 2.7mAP and 5.76mAR higher than YOLOv5m in small object detection.