A recent study published in the journal Opticathe researchers said. Produced by University College London Multi-contrast images available for thousands of complex scenarios to identify potentially dangerous objects (e.g. explosives) – this is made possible by combining different X-ray imaging techniques.
The new method could be useful for security screening and applications in the physical and biological sciences, and also uses an easily accessible machine learning process for classifying materials.
The method is particularly suitable for identifying objects with very similar elemental composition and can be used in airport security or in in-line scanning operations to inspect materials flagged as suspicious by initial rapid scans such as conventional X-ray systems..
Thomas Partridge, Research Team Leader, University College London
The new method successfully identified and detected explosives after nearly 4,000 scans of threatening and non-threatening materials hidden in bags and various objects, with only one false positive and a near-perfect detection rate of 99.68 percent for threatening cases.
Although more research is needed, the approach could also be useful in medical imaging: While traditional X-ray imaging struggles to distinguish healthy from diseased tissue, other research suggests that phase-contrast imaging may be able to capture textures that can be used to distinguish healthy from benign tissue..
Thomas Partridge, Research Team Leader, University College London
Unveiling the secrets of ingredients
X-ray attenuation describes the decrease in X-ray intensity as radiation passes through matter and is the basis of X-ray equipment found in airports and medical facilities. The new method combines traditional X-ray attenuation data at different X-ray energies with X-ray phase information consisting of refraction and dark-field channels to produce multi-contrast images.
Many explosives and common everyday items are composed primarily of carbon, hydrogen, nitrogen and oxygen and are difficult to distinguish using X-ray attenuation alone. The additional channels greatly enhance edges, material texture and grain, making it possible to distinguish between objects with very similar elemental composition..
Thomas Partridge, Research Team Leader, University College London
The study expands on the researchers' previous efforts to combine machine learning techniques with multi-contrast X-ray phase-enhanced imaging for the identification of threats with fewer explosives and benign objects.
To get even closer to real-world conditions, we have significantly increased the number of materials studied and scenarios imaged in the current experiments, and we have also developed a more efficient scanning system that allows us to tune the resolution by varying the scanning speed and applying phase contrast to the edge illumination.
In edge illumination, masks are placed in front of and behind the sample to provide the sub-pixel XR-ray “beamlets” required to increase the system's sensitivity to phase signals. This illumination approach can work with incoherent X-ray sources, expanding its usefulness, which is one of its main advantages.
The scientists used machine learning in a hierarchical design to separate cluttered objects before distinguishing between material types – as increasingly complex imaging settings necessitated more elaborate protocols – to rapidly identify materials based on their key distinguishing characteristics by detecting subtle changes in shape and texture.
Threat Detection
To test the new method, the scientists used 56 non-threat and 19 threat substances, all in three different thicknesses, concealed in various items that passengers bring in their carry-on luggage, including socks, brushes and face wipes.
In some instances, the researchers were able to demonstrate material discrimination as well as identification by utilizing all acquired contrast channels. In only one miss out of 313 hazardous situations was the signal from the combination of XR line contrasts able to be analyzed using deep learning, with very promising results.
According to the researchers, for this strategy to be commercially viable, the scanning speed needs to be increased through further system optimization, and the robustness of material identification needs to be tested using more extensive data sets.
The team's current research focus is combining this technology with 3D computed tomography scanning, which is being investigated for security applications as it can create precise 3D images of items.
Journal References:
Partridge, T. etc (2024) Multi-contrast X-ray discrimination of heterogeneous materials and their identification using a deep learning approach. OpticaSource: http://www.optica.org/en/default.aspx
Source: https://www.optica.org/
