Semiconductor Physics, Quantum Electronics & Optoelectronics, 24 (1), P. 90-99 (2021).

THz linear array scanner in application to the real-time imaging and convolutional neural network recognition
A.G. Golenkov, A.V. Shevchik-Shekera, M.Yu. Kovbasa, I.O. Lysiuk, M.V. Vuichyk, S.V. Korinets, S.G. Bunchuk, S.E. Dukhnin, V.P. Reva and F.F. Sizov

V. Lashkaryov Institute of Semiconductor Physics, National Academy of Sciences of Ukraine,
41, prospect Nauky, 03680 Kyiv, Ukraine

Abstract. Room temperature linear arrays (up to 160 detectors in array) from silicon metal- oxide-semiconductor field-effect transistors (Si-MOSFETs) have been designed for sub- THz (radiation frequency 140 GHz) close to real-time direct detection operation scanner to be used for detection and recognition of hidden objects. For this scanner, the optical system with aspherical lenses has been designed and manufactured. To estimate the quality of optical system and its resolution, the system modulation transfer function was applied. The scanner can perform real-time imaging with the spatial resolution better than 5 mm at the radiation frequency 140 GHz and contrast 0.5 for the moving object speed up to 200 mm/s and the depth of field 20 mm. The average dynamic range of real time imaging system with 160-detector linear array is close to 35 dB, when the sources with the output radiation power of 23 mW (IMPATT diodes) are used (scan speed 200 mm/s). For the system with 32-detector array, the dynamic range was about 48 dB and for the single-detector system with raster scanning 80 dB with lock-in amplifier. However, in the latter case for obtaining the image with the sizes 20×40 mm and step of 1 mm, the average scanning time close to 15 min is needed. Convolutional neural network was exploited for automatic detection and recognition of hidden items.

Keywords: linear sub-THz array, aspherical optics, scanning systems, convolutional neural network.

Full Text (PDF)

Back to Volume 24 N1

Creative Commons License
This work is licensed under a Creative Commons Attribution-NoDerivatives 4.0 International License.