Analysis Of Radiographic Images To Improve Radiological .

1y ago
88 Views
1 Downloads
8.22 MB
26 Pages
Last View : 1m ago
Last Download : 3m ago
Upload by : Francisco Tran
Transcription

Analysis of Radiographic Images to ImproveRadiological/Nuclear Threat Detection inCommercial Cargoand Experiences Working with DHS/CBPBrian S. Henderson([email protected])July 18, 2019CBP-ADEPT Workshop 2

SWWC1Space: Nuclear/radiological detection (SNM and RDDs) incommercial cargo2Problems: High cost of effective systems, poor overlap withother CBP missions, high false alarm rates3Solution: Better utilization of general purpose radiographysystems in conjunction with passive detection by leveragingexisting radiography data to characterize cargo streams4Results: Data analysis shows very high probability ofdetection of large class of nuclear/radiological threats atfalse positive rates of 2%5DoD TRL: Analysis/technique at TRL 2, but utilizes TRL 8/9hardware6Contact Information: [email protected] S. Henderson (MIT)July 18, 20192 / 11

Current Approaches to Nuclear Cargo Security Passive techniques Simple, low-cost Specific to nuclear materialImage Source: NNSA, Nevada Site Office Photo LibraryBrian S. Henderson (MIT)July 18, 20193 / 11

Current Approaches to Nuclear Cargo Security Passive techniques Simple, low-cost Specific to nuclear material Bremsstrahlung radiography More expensive in both timeand money Much more generalImage Source: Varian Medical SystemsBrian S. Henderson (MIT)July 18, 20193 / 11

Current Approaches to Nuclear Cargo Security Passive techniques Simple, low-cost Specific to nuclear material Bremsstrahlung radiography More expensive in both timeand money Much more general Active interrogation Typically specific to nuclearmaterial, very high cost Remains very much “on thedrawing board”Image Source: P.B. Rose, et al., Scientific Reports 6, 24388 (2016)Brian S. Henderson (MIT)July 18, 20193 / 11

What is the ideal solution? Speed: Must process a container in .1 minute Material sensitivity: In some way, must be sensitive to nuclearand radiological material Low false alarm rate: False positives are a key complaint ofport operators Easy Operation: System must be reliable, have a smallfootprint, and produce easy-to-understand alarms Ideally overlaps with other missions: Contraband/taxevasion detection, stowaways (both detection and dosesafety)Brian S. Henderson (MIT)July 18, 20194 / 11

Essential Approach This is the first analysis of a significant set of radiographicimages of cargo containers to assess the frequency ofobjects appearing similar to shielded nuclear/radiologicalthreats Utilizes a set of 120,000 images of 20 and 40 foot containerimages taken with a Rapiscan Eagle 6 MeV bremsstrahlungrail scanner at the Port of Rotterdam Essential approach: Model the appearance of relevant nuclear/radiologicalthreats in radiographs, characterized by their apparentsize/areal density Determine the frequency of objects of the relevantsizes/densities in the container stream (which amounts to afalse alarm rate using this technique in isolation)Brian S. Henderson (MIT)July 18, 20195 / 11

Some Sample Threats to ConsiderConsider the effective radius atthickness greater than 25 cm steelequivalent of a few example objects Bare U critical mass — 7.5 cm Assembled fission weapon — 12cm U pit shielded with 3 cm Pb on allsides — 10 cm Pu pit fully shielded againstneutron detection — &40 cmImage Source: Fetter, et al., Science & Global Security 1, 225 (1990) (TOP)Brian S. Henderson (MIT)July 18, 20196 / 11

Simulated Pu Device in a Container ImageDensity Threshold: 25.6 cm-steel equivalent250Vertical Position (cm)20150151001050Cargo Thickness (cm-steel equivalent)252005000100200300400500600Horizontal Position (cm)Brian S. Henderson (MIT)July 18, 20197 / 11

Density Distribution of Cargo10 -51.510.90.710.60.50.40.50.30.2Cumulative Pixel FrequencyPixel Frequency (normalized)0.80.100510152025030Thickness S (cm-steel equivalent)20 foot containers40 foot containersCharacterization of cargo stream properties at fine scaleBrian S. Henderson (MIT)July 18, 20198 / 11

Largest Object (by Radius) per 20-ft ContainerEffective false positive rate using this technique in isolationBrian S. Henderson (MIT)July 18, 20199 / 11

Image Analysis Conclusions This analysis shows that objects that appear like nuclearweapons occur in .2% of containers, several percent forother threat classes There is much to be learned by digging into this data andthere may be a significant opportunity to improvenuclear/radiological threat detection and inform othermissions Analysis of other data streams is critical, along with fusion ofmultiple data sources for containers CBP/DHS should seek to promote analyses of large datasets, and facilitate fusion of multiple sources. Much can begained with little or no development of noveltechnology/hardwareBrian S. Henderson (MIT)July 18, 201910 / 11

For More Information and Similar ApproachesUpcoming Publication of This WorkHenderson, B. S. “Analysis of the Frequency and Detectability ofObjects Resembling Nuclear/Radiological Threats in CommercialCargo”, In press. (Science and Global Security) (2018). Pre-print:arXiv:1901.03753.Unless otherwise noted on the slide, all images in this presentation arefrom this work.Related Machine-Learning Work Using Same Data for OtherCustoms/Security GoalsN. Jaccard, T. W. Rogers, and L. D. Griffin, in 2014 11th IEEE InternationalConference on Advanced Video and Signal Based Surveillance (AVSS)(2014) pp. 387–392.Brian S. Henderson (MIT)July 18, 201911 / 11

Extra SlidesBrian S. Henderson (MIT)July 18, 201912 / 11

The 9/11 Commission Act Mandate (2007)§1701. Container Scanning and SealsIN GENERAL.—A container that was loaded on a vessel in aforeign port shall not enter the United States (either directly or viaa foreign port) unless the container was scanned by nonintrusiveimaging equipment and radiation detection equipment at aforeign port before it was loaded on a vessel.Mandated for implementation in 2012, delayed 4 times sincethen, and no plan exists for meeting the next deadline in 2020Brian S. Henderson (MIT)July 18, 201913 / 11

Current Procedure at US PortsImage Source: Congressional Budget OfficeBrian S. Henderson (MIT)July 18, 201914 / 11

Previous Data on Cargo DensityImage Source: Descalle, et al. Analysis of Recent Manifests for Goods Imported through US Ports, UCRL-TR-225708.Brian S. Henderson (MIT)July 18, 201915 / 11

Image Data Set Parameters 120,000 images of 20 and 40 ft containers Dual energy 4 and 6 MeV bremsstrahlung beam 4 4 mm pixel size Penetration up to 30 cm steel equivalent 16-bit integrated transmittance measurement per pixel 20% empty containersBrian S. Henderson (MIT)July 18, 201916 / 11

The Rapiscan R60 Rail Scanner in RotterdamImage Source: Rapiscan SystemsBrian S. Henderson (MIT)July 18, 201917 / 11

Typical Container Image250202001530010400Steel Equivalent Depth (cm)Pixels ( 4 mm/pixel)100500560020040060080010001200Pixels ( 4 mm/pixel)Brian S. Henderson (MIT)July 18, 201918 / 11

Largest Object (by Radius) per 40-ft ContainerBrian S. Henderson (MIT)July 18, 201919 / 11

Largest Object (by Area) per 20-ft ContainerBrian S. Henderson (MIT)July 18, 201920 / 11

Largest Object (by Area) per 40-ft ContainerBrian S. Henderson (MIT)July 18, 201921 / 11

Mean Cargo Thickness Along Container Length1020 foot containers40 foot containersMean Cargo Thickness (cm-steel equivalent)987654321000.20.40.60.81Fraction of Container Length from Leading WallBrian S. Henderson (MIT)July 18, 201922 / 11

Mean Cargo Thickness Along Container Height15Mean Cargo Thickness (cm-steel equivalent)20 foot containers40 foot containers1050050100150200Height from Container Floor (cm)Brian S. Henderson (MIT)July 18, 201923 / 11

Data-Driven Single Pixel Uncertainty Estimate4Standard Deviation per Pixel (cm-steel equivalent)Measured2nd Degree Polynomial Fit3.532.521.510.50051015202530Reconstructed cm-steel EquivalentBrian S. Henderson (MIT)July 18, 201924 / 11

This is the first analysis of a significant set of radiographic images of cargo containers to assess the frequency of objects appearing similar to shielded nuclear/radiological threats Utilizes a set of 120,000 images of 20 and 40 foot container images taken with a Rapiscan Eagle 6