@article{MB47CC178,
title = "The Object Detector for Aerial Image Using High Resolution Feature Extractor and Attention Module",
journal = "The Journal of Korean Institute of Communications and Information Sciences",
year = "2023",
issn = "1226-4717",
doi = "10.7840/kics.2023.48.1.1",
author = "HaeMoon Kim, JongSik Ahn, Tae-Young Lee, Byungin Choi",
keywords = "Aerial Image, High Resolution Feature Map, Attention Network, Tiny Object Detection",
abstract = "Object detectors, such as YOLOv5, achieve high performance on datasets that consist of objects in everday scenes, like the COCO dataset. However, it shows poor performance in aerial images because the detectors did not consider the size of the objects. First, the aerial images contain very tiny objects and these objects are densely located in a image. Second, because of wide FOV, most of images has a lot of complex background information. It makes object detector very difficult to recognize object and background. In this paper, we propose an object detector that focuses on tiny objects with high resolution feature maps and attention network. We densely located in a image. Second, because of wide FOV, most of images has a lot of complex background information. It makes object detector very difficult to recognize object and background. In this paper, we propose an object detector that focuses on tiny objects with high resolution feature maps and attention network. We design SB network which is feature extractor through high resolution feature map. Also we adopted Triplet Attention to TA network for distinguish between objects and background. The proposed YOLOv5l-TA network and achieves  11.2% higher than YOLOv5l baseline network and 280%, 55%, 36.1%, 4.8% in  ,   ,  ,  metrics."
}