Machine learning allows a simple technique for recognizing concealed objects
DOI: 10.1063/10.0002928
Machine learning allows a simple technique for recognizing concealed objects lead image
The ability to find and recognize concealed objects is useful across fields, from determining the water distribution in biological substances to identifying hidden, potentially dangerous materials. Many materials are transparent to terahertz radiation, making it very useful for this application, but detecting terahertz radiation is challenging. To overcome this challenge, Limbacher et al. demonstrated a machine learning optical object recognition algorithm that can determine the shape of concealed objects.
After training the system, they found it can identify metal objects almost perfectly. Moreover, though the training process is done with the object in sight, the system can even classify objects that are covered behind a material, such as paper or polyethylene foil.
“This is actually very useful, as one may want to detect certain objects, but during training, there is no knowledge of the materials behind which the objects are hidden,” said author Benedikt Limbacher.
Central to this achievement is a spatially modulated terahertz signal. The modulation attenuates the light in certain places to generate a transmission mask, and compares the ratios of transmitted power with and without the mask to determine if the shape of the object blocking the beam matches that of a known object. This process allows it to conduct the majority of its calculations optically and in parallel.
“This not only makes the calculation inherently fast, but also makes 2D plane array detectors, such as cameras, obsolete,” Limbacher said. “Only a single pixel detector is required.”
Because of its flexibility and customizability, the group is optimistic of the technique’s use in quality control and security, where operators look out for certain materials and shapes.
Source: “Terahertz optical machine learning for object recognition,” by B. Limbacher, S. Schoenhuber, M. Wenclawiak, M. A. Kainz, A. M. Andrews, G. Strasser, J. Darmo, and K. Unterrainer, APL Photonics (2020). The article can be accessed at https://doi.org/10.1063/5.0029310 .
