THZ IMAGING OF HIDDEN OBJECTS AND THEIR CLASSIFICATION USING NEURAL NETWORKS

Ugnė Šilingaitė^{1, 2}, Ignas Grigelionis¹

¹ Center for Physical Sciences and Technology

² Vilnius University

[email protected]

Machine learning models have recently made various breakthroughs in solving computer vision tasks and have a high accuracy in object classification problems [1], thus can be applied to terahertz (THz) images. However, THz images are often plagued with the problem of low spatial resolution due to the limited hardware capabilities and long imaging times, thus there are no available datasets which would be suitable for neural network training. The goal of our work is to create a dataset containing THz images of different objects and use that dataset to train Convolution neural network models to detect and classify the objects in the dataset. Raster method was used to get 152 images, which were used for training and testing of the models.

In this work we trained a simple Convolutional neural network model for object classification and reached 95% overall accuracy. U-Net [2] model was also trained for object detection and reached 0,91 Dice coefficient on training data.

[1] HASSAN, Aya, Moatamed REFAAT, and Ashraf HEMEIDA. Image classification based deep learning: A Review. Aswan University Journal of Sciences and Technology [online]. 2022, 2(1), 11–35 [viewed 17 February 2025]. ISSN 2735-3095. Available from: doi:10.21608/aujst.2022.259887

[2] RONNEBERGER, Olaf, Philipp FISCHER, and Thomas BROX. U-Net: Convolutional Networks for Biomedical Image Segmentation. In: Lecture Notes in Computer Science [online]. Cham: Springer International Publishing, 2015, pp. 234–241 [viewed 17 February 2025]. ISBN 9783319245737. Available from: doi:10.1007/978-3-319-24574-4_28