데이터증강 모델 조합을 활용한 적성 전차에 대한 객체탐지 성능 향상 연구

• Published in: 한국CDE학회 논문집 Vol. 27; no. 2; pp. 148 - 159
• Main Authors: 황교성(Kyoseong Hwang), 마정목(Jungmok Ma)
• Format: Journal Article
• Language: Korean
• Published: (사)한국CDE학회 01.06.2022; 한국CDE학회
• Subjects: 기계공학; SinGAN; Object Detection; Data Augmentation; Mosaic Augmentation; RandAugment YOLOv5
• ISSN: 2508-4003; 2508-402X
• DOI: 10.7315/CDE.2022.148

With the development of artificial intelligence technology, it is continuously being applied in the defense field. In particular, deep learning-based object detection technology is treated as a groundbreaking technology in the field of defense surveillance and reconnaissance. However, there are limitations to use deep learning-based object detection technology because it is hard to get enough image data of enemy weapon systems. To overcome this challenge, this paper studies the improvement of object detection performance for enemy tanks using the combination of data augmentation models. Experiment results show that the combination of selected data augmentation models improves object detection performance(especially SinGAN model is effective). The result indicates that the data augmentation in the field of defense surveillance and reconnaissance needs to be studied since the result of combining all data augmentation models would not necessarily be good. KCI Citation Count: 9