Automated Threat Recognition (ATR) Algorithms for Explosion Detection Systems
ADSA08

The final report for this workshop available at:

https://myfiles.neu.edu/groups/ALERT/strategic_studies/ADSA08_final_report.pdf.

ADSA08 focused on automated threat recognition (ATR) algorithms for explosion detection systems was held at Northeastern University in Boston on October 24-25, 2012.

The topic of ATR was chosen for the workshop in order to support the Department of Homeland Security’s (DHS) objective of improving the detection performance of existing technologies. Improved detection performance is defined as increased probability of detection, decreased probability of false alarms, lower detected threat mass and an increase in the number of types of explosives detected.

The topics that were addressed at the workshop are as follows:

  1. ATR for:
    1. CT-based EDS
    2. Whole body imaging (WBI) & Advanced Imaging Technology (AIT)
    3. Carry-on baggage inspection
    4. Cargo
    5. Trace
    6. Fused systems
    7. Risk based screening
    8. Behavioral detection
    9. Detection explosives implanted in a passenger’s body
    10. XBS dose
    11. Accelerating deployment of third party advances
    12. Deterrence

 

The presentations and discussions concentrated on imaging devices such as CT-based EDS and Advanced Imaging Technology.

The workshop was successful fostering interaction between third parties and vendors, reducing barriers to their working together, now and in the future.  It also directly led to increased third party involvement in the development of advanced ATR algorithms. This conclusion is based on anecdotal evidence of the number of third parties engaging in discussions with vendors during the workshop and the editors’ knowledge of third parties consulting for the vendors.

Workshop Outcomes

  • For an imaging device:
    • ATR should be defined as an operation with images as input and a yes/no decision on the presence of a threat as an output
    • ATR should include the following steps: segmentation, feature extraction, correction for device imperfections and classification.
    • It would be very difficult for a third party to develop, without direct assistance from a vendor, an ATR for a deployed explosive detection device (e.g., an EDS) for the following reasons.
      • Detection requirements are classified
      • Data from deployed equipment are SSI or classified, and are under export control
      • There is no publicly available set of images that are representative of challenging ATR problems for explosive detection systems.
      • The business interests of the vendors should be protected
      • DHS/TSA policies do not allow TSL to test components (e.g., an ATR) separate from a complete scanner.
    • Third parties can make advances to ATR by working with data and requirements that are in the public domain. This task could be accomplished through the following steps.
      • Detect a set of benign objects such as peanut butter and rubber sheets.
      • Write detection requirements based on these benign objects
      • Scan these objects on an equivalent device in a related field. For example, for x-ray based EDS, scan on a medical CT scanner.
      • Provide an environment in which third parties, industry and Government can interact
    • The following topics should be considered in detail in the future.
      • Developing and testing ATRs with few training and test samples.
      • Developing metrics for improved performance when the confidence intervals for tests of PD and PFA are large due to small data sets.
      • Funding for researchers from DHS, TSA, government laboratories, and industry.
      • Incentives from the TSA for vendors to deploy equipment with improved detection performance. These incentives will lead to the deployment of advanced ATR algorithms.
      • Developing ATRs with support for risk-based screening, deterrence and the human in the loop.