Lithuanian Journal of Physics article / Lietuvos fizikos žurnalo straipsnis

CLASSIFICATION OF CONCEALED OBJECTS USING TERAHERTZ IMAGING AND ARTIFICIAL NEURAL NETWORKS
Ugnė Šilingaitė and Ignas Grigelionis
Center for Physical Sciences and Technology, Saulėtekio 3, 10257 Vilnius, Lithuania
Email: ugne.silingaite@ftmc.lt

Received 29 November 2025; accepted 3 December 2025

Imaging in the terahertz frequency band is applied in a number of fields, such as security, medical or quality control. However, a low resolution or distortions of the images hinder the identification or recognition of the objects. To cope with the processing of visual information, artificial neural networks are broadly employed. In this work, the monochromatic radiation of 253 GHz was used to collect the image set of the investigated objects either in the air or covered with a packing material. Such a set was later used to train convolutional and generative adversarial neural networks poised for three tasks: (i) the classification of objects; (ii) the enhancement of image resolution; (iii) the identification of cover material. The obtained results demonstrated that the packaging materials were identified with an accuracy of 83.33%, while the investigated objects were classified with an accuracy of 89.42%. The PSNR metric of images with improved resolution reached up to 22.44 dB. The optical properties such as refractive indices and absorption coefficients of the packaging materials were also defined using terahertz time-domain spectroscopy, and it was found that the accuracy of object and material classification in general does not depend on the physical properties and type of a package.
Keywords: terahertz imaging, neural networks, image processing

PASLĖPTŲ OBJEKTŲ KLASIFIKACIJA NAUDOJANT TERAHERCINĮ VAIZDINIMĄ IR DIRBTINIUS NEURONINIUS TINKLUS
Ugnė Šilingaitė, Ignas Grigelionis

Fizinių ir technologijos mokslų centras, Vilnius, Lietuva

Terahercinį (THz, bangos ilgiai 0,3–3 mm) vaizdinimą galima taikyti daugelyje sričių, pavyzdžiui, medicinoje, kokybės kontrolės ar saugumo srityse, nes šio dažnio bangos gali prasiskverbti pro įvairias dielektrines medžiagas. Tačiau THz vaizdams būdinga maža vaizdo raiška ir didelis triukšmas, todėl sudėtinga panaudoti šį bangos ilgį objektų atpažinimui ir klasifikavimui. Šią problemą galima spręsti pasitelkiant dirbtinius neuroninius tinklus, gebančius apdoroti vizualinę informaciją. Šiame darbe objektų vaizdų duomenų bazė buvo surinkta naudojant monochromatinę 253 GHz dažnio spinduliuotę. Vaizdinami objektai buvo arba atviri, arba uždengti įvairiomis pakavimo medžiagomis. Surinkta duomenų bazė buvo naudojama skirtingiems neuroniniams tinklams apmokyti, kurie atlieka tris užduotis: (i) klasifikuoja objektus; (ii) gerina vaizdo kokybę; (iii) klasifikuoja objektus dengiančias pakavimo medžiagas. Gauti rezultatai parodė, kad pakavimo medžiagos yra klasifikuojamos 83,33 % tikslumu, o individualūs objektai – 89,42 % tikslumu. Vaizdų raiškos gerinimas priklauso nuo pakavimo medžiagos, o geriausia gauta signalo ir triukšmo santykio vertė siekia 22,44 dB. Naudojant THz-TDS sistemą nustatytos pakavimo medžiagų optinės savybės, tačiau tyrimo metu paaiškėjo, kad jos nekoreliuoja su objektų ir medžiagų klasifikacijos rezultatais.

References / Nuorodos

[1] A. Leitenstorfer, A.S. Moskalenko, T. Kampfrath, J. Kono, E. Castro-Camus, K. Peng, N. Qureshi, D. Turchinovich, K. Tanaka, and A. Markelz, The 2023 terahertz science and technology roadmap, J. Phys. D 56, 223001 (2023),
https://doi.org/10.1088/1361-6463/acbe4c
[2] X. Li, J. Li, Y. Li, A. Ozcan, and M. Jarrahi, High-throughput terahertz imaging: progress and challenges, Light Sci. Appl. 12, 233 (2023),
https://doi.org/10.1038/s41377-023-01278-0
[3] R. Piesiewicz, C. Jansen, S. Wietzke, D. Mittleman, M. Koch, and T. Kürner, Properties of building and plastic materials in the THz range, Int. J. Infrared Milli. Waves 28, 363–371 (2007),
https://doi.org/10.1007/s10762-007-9217-9
[4] A. Nakanishi and H. Satozono, Terahertz optical properties of wood-plastic composites, Appl. Opt. 59, 904–909 (2020),
https://doi.org/10.1364/AO.379758
[5] A. Prokscha, F. Sheikh, K. Kolpatzeck, Y. Zantah, D. Lessy, J. Balzer, A. Czylwik, and T. Kaiser, Terahertz insights into fabric look-through, in: Proceedings of 2024 15th German Microwave Conference (IEEE, 2024) pp. 241–244,
https://doi.org/10.23919/GeMiC59120.2024.10485292
[6] T.S. Hartwick, D.T. Hodges, D.H. Barker, and F.B. Foote, Far infrared imagery, Appl. Opt. 15, 1919–1922 (1976),
https://doi.org/10.1364/AO.15.001919
[7] B.B. Hu and M.C. Nuss, Imaging with terahertz waves, Opt. Lett. 20, 1716–1719 (1995),
https://doi.org/10.1364/OL.20.001716
[8] A.G. Davies, A.D. Burnett, W. Fan, E.H. Linfield, and J.E. Cunningham, Terahertz spectroscopy of explosives and drugs, Materials Today 11, 18–26 (2008),
https://doi.org/10.1016/S1369-7021(08)70016-6
[9] T. Mohr, S. Breuer, G. Giuliani, and W. Elsäßer, Two-dimensional tomographic terahertz imaging by homodyne self-mixing, Opt. Express 23, 27221-27229 (2015),
https://doi.org/10.1364/OE.23.027221
[10] K. Komatsu, T. Iwamoto, H. Ito, and H. Saitoh, THz gas sensing using terahertz time-domain spectroscopy with ceramic architecture, ACS Omega 7, 30768–30772 (2022),
https://doi.org/10.1021/acsomega.2c01635
[11] Y. Peng, C. Shi, X. Wu, Y. Zhu, and S. Zhuang, Terahertz imaging and spectroscopy in cancer diagnostics: A technical review, BME Front. 25, 2547609 (2022),
https://doi.org/10.34133/2020/2547609
[12] K. Ikamas, D.B. But, A. Cesiul, C. Kołaciński, T. Lisauskas, W. Knap, and A. Lisauskas, All-electronic emitter-detector pairs for 250 GHz in silicon, Sensors 21, 5795 (2021),
https://doi.org/10.3390/s21175795
[13] A. Siemion, L. Minkevičius, L. Qi and G. Valušis, Spatial filtering based terahertz imaging of low absorbing objects, Opt. Lasers Eng. 139, 106476 (2021),
https://doi.org/10.1016/j.optlaseng.2020.106476
[14] Y. Freund and R.E. Schapire, Large margin classification using the perceptron algorithm, Mach. Learn. 37, 277–296 (1999),
https://doi.org/10.1023/A:1007662407062
[15] A. Derry, M. Krzywinski, and N. Altman, Convolutional neural networks, Nat. Methods 20(9), 1269–1270 (2023),
https://doi.org/10.1038/s41592-023-01973-1
[16] D. Liang, F. Xue, and L. Li, Active terahertz imaging dataset for concealed object detection (2021),
https://doi.org/10.48550/arXiv.2105.03677
[17] Y. Li, W. Hu, X. Zhang, Z. Xu, J. Ni, and L.P. Ligthart, Adaptive terahertz image super-resolution with adjustable convolutional neural network, Opt. Express 28, 22200 (2020),
https://doi.org/10.1364/OE.394943
[18] Z. Long, T. Wang, C. You, Z. Yang, K. Wang, and J. Liu, Terahertz image super-resolution based on a deep convolutional neural network, Appl. Opt., 58, 2731 (2019),
https://doi.org/10.1364/AO.58.002731
[19] O. Kupyn, V. Budzan, M. Mykhailych, D. Mishkin, and J. Matas, DeblurGAN: Blind motion deblurring using conditional adversarial networks, in: 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 8183 (IEEE, 2018),
https://doi.org/10.48550/arXiv.1711.07064
https://doi.org/10.1109/CVPR.2018.00854
[20] A. Thakur, Generative adversarial networks, Int. J. Res. Appl. Sci. Eng. Technol. 9, 2307–2325 (2021),
https://doi.org/10.22214/ijraset.2021.37723