Ryoya Mitsuhashi, Kosuke Murate, Seiji Niijima, Toshinari Horiuchi, Kodo Kawase,
 |
| (a) Tags used for real-time identification. Natural leather was used as the shielding material. Tags are shown for Chemicals A, B, and C (from the right) and were attached to the shielding material. Tags were moved in the direction of the arrow. (b) Images of the detection beams of the multi-wavelength is-TPG without tag insertion. For this measurement, we selected three wavelengths to match the absorption peaks of each tag. (c) The intensity spectrum of each tag measured with a conventional is-TPG. The symbols show the intensities of the three frequencies selected in this study; the red and blue symbols represent the transmission and absorption frequencies, respectively. (d) We took screenshots showing the timing of the measurement of each tag’s position from the video of tag sweeping. The names of tags identified by convolutional neural network (CNN), and the identification probability, are shown in the upper right corner. |
In recent years, there has been great interest in chipless radio-frequency identification (RFID) devices that work in the terahertz (THz) frequency range. Despite advances in RFID technology, its practical use in the THz range has yet to be realized, due to cost and detection accuracy issues associated with shielding materials. In this study, we propose two types of low-cost THz-tags; one is based on the thickness variation of coated polyethylene and the other on the fingerprint spectra of reagents. In the proposed approach, machine learning, specifically a deep-learning method, is used for high-precision tag identification even with weak signals, or when the spectrum is disturbed by passing through shielding materials. We achieved almost 100% identification accuracy despite using an inexpensive tag placed under thick shielding materials with an attenuation rate of about −50 dB. Furthermore, real-time tag identification was demonstrated by combining a multiwavelength injection-seeded THz parametric generator and a convolutional neural network.
© 2020 Optical Society of America under the terms of the OSA Open Access Publishing Agreement
No comments:
Post a Comment