Material analysis technology based on terahertz imaging using deep learning

Abstract

Security checks using terahertz waves are now being commercialized. Existing security checks that rely on Xrays can see only the shapes of objects, so it is necessary to open and check items if there are suspicious objects within. In this study, I develop a technique to figure out not only the shapes of objects but also their constituent materials by taking advantage of how the absorption spectrum of the terahertz wave differs from material to material. To do this, I obtain multi-channel terahertz images with various frequency terahertz waves and analyze the constituent materials by changing the transmittance of terahertz waves. Deep learning is applied here to process a large amount of calculations in a short time and determine the type of constituent material. The 2D convolutional neural network, which is commonly used in image classification, is judged only by the shape of an object. However, in this study, a 3D convolutional neural network is used to scan the z-axis and identify the characteristics of the absorption spectrum reflected in terahertz images. This allows us to accurately judge objects that were previously difficult to judge by their shapes alone in security checks based on material information, and therefore this technology is expected to lead to a breakthrough in the field of security checks.