Researchers at University College London Developed a Deep Learning-based Method to X-Ray Luggage to Detect Explosives

The development of phase-based techniques has accelerated the pace of X-ray imaging. Dark-field images are sensitive to inhomogeneities on a length scale below the system’s spatial resolution, and phase contrast images are improved for detailed visibility. A new technique for X-raying luggage to find trace levels of explosives was developed by a team of researchers from University College London, Nylers Ltd., and XPCI Technology Ltd. They demonstrated how dark-field produces a texture specific to the substance being photographed and how combining it with traditional attenuation improves the ability to distinguish amongst threat materials. They have also published their work in Nature Communications journal, which involves adapting a conventional X-ray detector and using a deep-learning application to better detect hazardous chemicals in luggage.

Additionally, their research demonstrates that lingering misunderstandings can be cleared up by using the different energy dependences of the dark-field and attenuation signals. Additionally, two proof-of-concept experiments show that dark-field texture is suitable for identification using machine learning techniques. The performance suffered when identical methods were applied to datasets from which the darkfield images had been eliminated. Prior studies have demonstrated that the type of material significantly impacts the micro bends produced when X-rays interact with it. The researchers aimed to leverage these bends to build a precise X-ray system.

The first modification the researchers made to existing X-ray equipment was adding a box containing masks that are metal sheets with tiny holes punched through them. The purpose of the masks is to divide the X-ray beam into numerous smaller beams. A deep-learning AI application was then fed the scan findings from various objects containing embedded explosive ingredients. The goal was to educate the machine on how to perceive the appearance of such materials’ minute bends. After the machine had been trained, they tested its abilities by scanning other items for embedded bombs. Under lab conditions, the researchers discovered that their device achieved 100% accuracy.

The device successfully identified bends as small as one microradian, or one 20,000th of a degree. However, there is still scope for further investigation because the research was done on a smaller scale. The overall results point to the possibility of combining the usage of deep neural networks and dark fields in applications outside security. The team believes its method may be slightly modified for use in other applications, such as medicine, in addition to being helpful to transportation security personnel. They think it may be trained to locate malignancies that are too small to be detected by standard testing equipment and to find minute flaws in surfaces of buildings or airplanes.

This Article is written as a research summary article by Marktechpost Staff based on the research paper 'Enhanced detection of threat materials by dark-field x-ray imaging combined with deep neural networks'. All Credit For This Research Goes To Researchers on This Project. Check out the paper and reference article.

Please Don't Forget To Join Our ML Subreddit

Khushboo Gupta is a consulting intern at MarktechPost. She is currently pursuing her B.Tech from the Indian Institute of Technology(IIT), Goa. She is passionate about the fields of Machine Learning, Natural Language Processing and Web Development. She enjoys learning more about the technical field by participating in several challenges.