Full Article: PDF
Scientific Object Identifier: http://s-o-i.org/1.1/TAS-03-95-49
DOI: https://dx.doi.org/10.15863/TAS.2021.03.95.49
Language: English
Citation: Danso, S. A., et al. (2021). Generative Adversarial Network Aided Security Check with Terahertz Images Using Improved YOLOv3. ISJ Theoretical & Applied Science, 03 (95), 301-309. Soi: http://s-o-i.org/1.1/TAS-03-95-49 Doi: https://dx.doi.org/10.15863/TAS.2021.03.95.49 |
Pages: 301-309
Published: 30.03.2021
Abstract: Terahertz imaging technology has the advantages of rapid imaging, strong penetration, harmless to the human body hence widely used in a variety of security environments and has become an alternative technology for X-ray imaging. In this paper, a GAN network-assisted deep learning method is proposed to detect whether there are illegal objects in terahertz images. Most importantly, our GAN model is used to improve the low resolutions of terahertz image and video. First, the GAN network reconstructs the blurred terahertz image, and then we use the optimized YOLOv3 object detection network to detect. The experimental results show that with the aid of GAN to blur image reconstruction, the accuracy of object detection is improved by 7.49%. On the YOLOv3 detection network, we added additional YOLO heads, which help to improve the ability of the network to detect objects of different sizes. Compared with the original YOLOv3 model, the improved model improves the detection performance by 10.96%. On the quantitative enhancement indexes (PSNR and SSIM), we attain a significant percentage increase.
Key words: Terahertz image, Terahertz technology, object detection, GAN model, optimized YOLOv3.
|
