Industries / Transport and vehicles / Maritime and shipping / Ports and terminals
Customs and inspection
Customs and inspection is the checking of cargo at the port by officers: choosing which containers to look at, scanning them with X-rays and radiation detectors, and opening the few that look wrong.
5
Milestones achieved, from 2002 to 2019.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
2002
Containers checked before loading
Within months of the September 2001 attacks, the US Customs Service created the Container Security Initiative (CSI), announced in January 2002. US officers posted at foreign ports work with local customs to pick high-risk containers with automated targeting tools and screen them before they are loaded for the United States, using large X-ray and gamma-ray machines and radiation detectors. By the count of US Customs and Border Protection (CBP), its 61 CSI ports now prescreen over 80% of the container cargo shipped to the US by sea.
Credit
US Customs Service, now CBP
2004
A security plan per terminal
The International Ship and Port Facility Security (ISPS) Code came into force on 1 July 2004 under chapter XI-2 of SOLAS, the main international treaty on the safety of ships. It requires governments, port authorities and shipping companies to appoint security officers, including a Port Facility Security Officer for each port facility, who assesses threats and writes and runs a security plan.
Credit
International Maritime Organization (IMO)
2012
Radiation gates at busy ports
The SAFE Port Act of 2006 requires every container entering the US by ship through the 22 busiest container seaports to be screened for radiation. In January 2013 the Office of Inspector General (OIG) of the Department of Homeland Security reported that, by the count of DHS's nuclear detection office, 444 radiation portal monitors (drive-through detector gates) at seaports were screening 99% of incoming container cargo as of August 2012. CBP reported more than 679 million containers screened and more than 2.8 million alarms since the program began, and about $623 million was spent from 2002 to 2011. The OIG also found monitors that had been installed and then not used for several years.
LessonWithout records of why each gate was placed where it was, auditors could not tell whether idle gates had been put in the wrong place.
Credit
CBP and the Domestic Nuclear Detection Office; audit by DHS OIG
2012
Full scanning target missed
The US set July 2012 as the target for scanning every US-bound container at foreign ports before loading. The Government Accountability Office (GAO), Congress's auditor, reported that DHS officials expected most foreign ports could not meet it. The Secretary of Homeland Security then issued three two-year extensions, to July 2018, and GAO said DHS "has not yet identified a viable solution."
LessonScanning every container abroad proved far harder than screening the few picked as high risk.
Credit
reported by GAO
2019
One file format for scanners
Under the World Customs Organization (WCO), four scanner makers, L3 (now part of Leidos), Nuctech, Rapiscan and Smiths Detection, developed the Unified File Format (UFF), a non-proprietary format for images from the high-energy scanners that inspect trucks, containers and rail wagons. The WCO released UFF 2.0, the first operational version, in July 2019.
Credit
WCO Technical Experts Group on Non-Intrusive Inspection
Watch for
AI reads India's container scans
Officers of India's Central Board of Indirect Taxes and Customs (CBIC) wrote in WCO News that they built, in house, an AI system at the National Customs Targeting Centre that analyzes X-ray images of containers at Indian seaports and flags anomalies. Its models classify products, detect objects (with YOLOv7) and predict container weight. They say detection accuracy has improved and give finds such as cigarettes hidden in cargo declared as air fresheners, but publish no accuracy or hit-rate figures.
Achieved whenan outside party publishes the system's hit rate and false-alarm rate on a stated set of containers.
Claimed by
CBIC, March 2026
A new scanner image standard
The WCO says draft specifications for UFF 3.0.0 went to its Policy Commission in December 2024 and were published as a provisional standard, and that Dutch Customs chose UFF as the format for its library of scanned container images, which it uses to build automated threat detection.
Achieved whenthe WCO Council approves UFF 3.0.0 and scanners from at least two makers write it in routine use.
Claimed by
WCO Secretariat, June 2025
Open
01
An open reader for UFF
A free, open-source library and viewer that reads UFF scanner files, shows the image and its metadata, and converts the image to formats used to train AI models.
The briefCredit · unclaimed02
Public scans of harmless cargo
A labeled public set of simulated container X-ray images of ordinary cargo only, each tagged with its product group under the Harmonized System (HS, the world's customs code for goods), for testing whether an AI can tell if a scan matches the goods declared.
The briefCredit · unclaimed
Open 01 · the brief
An open reader for UFF
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A free, open-source library and viewer that reads UFF scanner files, shows the image and its metadata, and converts the image to formats used to train AI models.
Why it is still open
India's customs officers wrote in March 2026 that "the images generated by the container scanning systems are shared and processed in other formats," and listed deploying UFF as work still to do. Searches of GitHub on 18 November 2026 found no public UFF reader; the one public cargo X-ray toolkit found reads files from VACIS scanners.
Built with what exists
The UFF 3.0.0 technical specification, which the WCO posts as a free Word file; it defines each case as an XML file with 16-bit TIFF X-ray images, and includes the XML schema and an example file. Python, with the lxml and tifffile libraries; an AI coding agent that writes code against a test suite (see the Note "Code no person read").
First step
Download the specification, extract the schema and example file, and list the fields a minimal reader must handle.
Done looks like
The library validates the WCO's example file against the schema, round-trips files it writes itself, and opens UFF files from at least two scanner makers, with a customs administration or maker confirming that the images and metadata match its own viewer.
Open 02 · the brief
Public scans of harmless cargo
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A labeled public set of simulated container X-ray images of ordinary cargo only, each tagged with its product group under the Harmonized System (HS, the world's customs code for goods), for testing whether an AI can tell if a scan matches the goods declared.
Why it is still open
A 2016 review of the field by Thomas Rogers and colleagues called the lack of open datasets "the main problem" in cargo image analysis and proposed sets of benign, non-sensitive items labeled with HS codes. It reported that the best published method for matching a scan to its declared HS code reached 48% accuracy at a 5% false-positive rate. The one public synthetic cargo set found, CargoX (64,000 images, posted on the Open Science Framework with a 2022 paper), labels only knives placed into cargo scans, not the declared goods. The customs image libraries named in WCO articles, India's and Dutch Customs', are held in house.
Built with what exists
The gVirtualXRay simulator (Python package gvxr), which computes X-ray images from 3D models of objects, so no radiation source is used; 3D models of common goods; an object-detection model such as YOLO for a baseline score.
First step
Simulate one 20-foot container of one product, such as stacked tires, and compare it with a published real scan.
Done looks like
5,000 simulated scans across 20 HS chapters, released under an open license with a baseline model and its accuracy, and 100 random images rated for realism by a customs image analyst.