Identifying threatening drone based on deep learning with SWIR camera
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
Small drones of various sizes are used in numerous fields, including commerce, reconnaissance, and offensive attacks. Major facilities such as security areas of port, power, and offshore plants urgently need to develop solutions for detecting drones as an active countermeasure against small drone attacks because small drones used for military and terrorism pose a significant threat. It is not easy to detect various drones such as invasive or threatening ones, though recent developments have made it possible to detect them using three-dimensional radar. Therefore, this paper develops threatening drone identification system, which consists of two components: One is a software component for identifying threatening drones among various ones and the other is a hardware component for the system. The former uses well-known YOLO(You Look Only Once) (v7) model and the latter comprises a PC for running the model and an SWIR (Short-Wave InfraRed) camera for surveillance. Datasets for training and evaluation are constructed by hand from air-borne videos taken drones including threating one and is labelled by two types: normal and threatening. The datasets are comprised of 3,992 color images and 4,410 thermal images, which are trained separately. Through experiments, we have shown that mAP@.5 and mAP@.95 are 0.999 and 0.753 (0.999 and 0.760) for color images (for thermal images), respectively. Consequently the proposed system is helpful in identifying threating drones.
Abstract
Small drones of various sizes are used in numerous fields, including commerce, reconnaissance, and offensive attacks. Major facilities such as security areas of port, power, and offshore plants urgently need to develop solutions for detecting drones as an active countermeasure against small drone attacks because small drones used for military and terrorism pose a significant threat. It is not easy to detect various drones such as invasive or threatening ones, though recent developments have made it possible to detect them using three-dimensional radar. Therefore, this paper develops threatening drone identification system, which consists of two components: One is a software component for identifying threatening drones among various ones and the other is a hardware component for the system. The former uses well-known YOLO(You Look Only Once) (v7) model and the latter comprises a PC for running the model and an SWIR (Short-Wave InfraRed) camera for surveillance. Datasets for training and evaluation are constructed by hand from air-borne videos taken drones including threating one and is labelled by two types: normal and threatening. The datasets are comprised of 3,992 color images and 4,410 thermal images, which are trained separately. Through experiments, we have shown that mAP@.5 and mAP@.95 are 0.999 and 0.753 (0.999 and 0.760) for color images (for thermal images), respectively. Consequently the proposed system is helpful in identifying threating drones.
참고문헌 (18)
- H. -G. Kim, "Analysis on patent trends for industry of un-manned aerial vehicle", Proceedings of the 2017 Fall Con-ference of the Korean Entertainment Industry Association, pp.90~93 (2017)
- S. H. Choi, "Recent R&D trends of anti-drone technologies", Electronics and Telecommunications Trends, 33(3), pp.78~88 (2018)
- 이동현, "심층 컨벌루션 신경망 기반의 실시간 드론 탐지 알고리즘", 로봇학회 논문지, 12(4), pp.425~431 (2017), 한국로봇학회[DOI]
- Y. LeCun, "Backpropagation ap-plied to handwritten zip code recognition", Neural Compu-tation, 1(4), pp.541~551 (1989)
- J. Redmon, "You only look once : Unified, real-time object detection", Pro-ceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp.779~788 (2016)
- J. Redmon, "YOLO9000: Better, faster, stronger", Proceedings of the IEEE Conference on Com-puter Vision and Pattern Recognition, pp.6517~6525 (2017)
- J. Redmon, "YOLOv3: An incremental im-provement", arXiv:1804.02767, ~ (2018)
- A. Bochkovskiy, "YOLOv4 : Optical speed and accuracy of object detection", arXiv:2004.10934, ~ (2020)
- C. -Y. Wang, "YOLOv7 : Trainable bag-of-freebies sets new state-of-the-art for real-Time object detectors", arXiv : 2207. 02696, ~ (2022)
- 강해영, "환경 소음 제거를 통한 범용적인 드론 음향 탐지 구현", 정보보호학회논문지, 32(5), pp.881~890 (2022), 한국정보보호학회[DOI]
- 이건영, "영상 기반 드론 식별 및 위치 표시", 전기학회논문지, 69(2), pp.331~336 (2020), 대한전기학회
- P. Nguyen, "Cost-effective and passive RF-based drone presence detection and characterization", GetMobile : Mobile Computing and Communications, 21(4), pp.30~34 (2017)
- 박호민, "Deep learning–based drone detection with SWIR cameras", 한국마린엔지니어링학회지, 44(6), pp.500~506 (2020), 한국마린엔지니어링학회[DOI]
- I. Alnujaim, "Generative adversarial networks for classification of micro-Doppler signatures of human activity", IEEE Geoscience and Remote Sensing Letters, 17(3), pp.396~400 (2020)
- Y. Son, "Rocking drones with intentional sound noise on gyroscopic sensors", Proceedings of the 24th USENIX security symposium, pp.881~896 (2015)
- Drone Identifica-tion and Tracking in Denmark, "K. Jensen", University of Southern Denmark, ~, pp.Technical Report TR-2016-2 (2016)
- Y. Chen, "A deep learning approach to drone monitoring", Proceedings of APSIPA Annual Summit and Conference, pp.686~591 (2017)
- "GitHub – heartexlabs/labelImg: Video/Image Labeling and Annotation Too"[URL]