On International Customs Day 2016, the Secretary General of WCO, Kunio Mikuriya, reminded us all that information and communications technology (ICT) is everywhere in today’s Customs workplace, and that ICT has transformed the way Customs and governments operate.

To meet the challenge of facilitating trade through more effective and streamlined control of merchandise, Customs authorities must take advantage of technologies, and more particularly of non-intrusive inspection (NII) solutions.

Over the last decades, Customs have stayed at the forefront of this key technology. Thousands of advanced NII systems – commonly X-ray transmission systems – have been installed at Customs security checkpoints around the world, enabling them to expedite the movement, release and clearance of internationally traded goods, while improving their fraud detection capacities.

Today, with the development of automatic detection tools, Customs’ capacity to detect fraud could take a giant leap.

Amazing tools as they are, NII systems need to be operated by highly skilled officers. Although great strides have been made in image processing over the years, there is still a scarcity of image analysts in many countries.

Human resources constraints is a long-standing problem which affects many Customs authorities. NII literacy is truly hard to earn in the short term. To train an image analyst, even a naturally talented one, requires a great deal of time to be given. One also has to take into consideration the need for constant training to keep pace with the booming development of new NII systems.

The good news is that technologies that will improve image interpretation are here, and now. Scanning technology providers are working on algorithms that will enable machines to recognize objects. Automatic detection tools are commonly known by the acronym ‘ATR,’ which means automatic threat recognition, or assisted target recognition. Big Data, cloud computing, machine learning, and advanced data analytics have been seamlessly integrated together to develop such solutions.

NII equipment is usually used in a standalone manner. An X-ray image collected from a scan is, in most cases, only used once at the checkpoint level. Field officers are required to analyse the X-ray image and assess whether it reveals the presence of suspicious items. Once a decision is made, X-ray images are often archived onsite, or stored in a remote server. They are usually not used again. In today’s information age, such a practice is seen by professionals as a huge loss.

Supported by a secure information technology (IT) system, Customs officials can now collect, manage and process tremendous volumes of historic scanning images, and build up a massive reference database which can be used for training purposes, as well as for building ATR algorithms.

ATR technologies are based on machine learning, which is the development of algorithms that learn from experience. Algorithms are developed to find specific patterns in X-ray scans of all sorts. The algorithm analyses pictures based on their shape, density, texture, and even atomic numbers to formulate certain pattern recognition rules, which can be applied later to other image sets, irrespective of the NII system used to produce the images.

The nature of the goods can be automatically identified and designated with distinctive colours according to the Harmonized System (HS) code of the commodity. Based on its unique pattern, ‘milk’ can be easily distinguished from ‘beverages’ for example. By comparing the image with the data contained in the cargo manifest, a machine would be able to assist field officers in verifying whether the data and image received from the scanner match.

Specific knowledge can be used to improve the application. For example, algorithms can be specifically fine-tuned to automatically distinguish consignments which are frequently traded. Filtering mechanisms can be put in place to free the operators from repetitive interpretation of similar scanned images of frequent goods.

The machine can learn also to detect certain items of interest such as cigarettes, bottled liquor, radioactive material, solid waste, or even artillery or weapons. The list of target items can be expanded and customized to accommodate specific needs.

By comparing the features of different parts of the image, an algorithm can also be built to pinpoint and highlight anomalies, for example a hidden compartment. Such technology also enables ‘ISO non-empty containers’ to be revealed, in order to detect contraband and illegal goods, forgotten or intentionally left in an ISO container, including differences in a load.

The machine learns from human interactions, and constantly expands its analytical capacities to better categorize, classify and detect goods. In other words, it sharpens its skills to become an expert in image interpretation through constant self-development.

ATR may also serve as a deterrent for the corruption of, or collusion between, officers and smugglers. Whatever the final decision regarding the release of a shipment, the system will record any hit in a database for future auditing, or even report it to the central command centre in real time.

ATR application is capable of memorizing details from tons of screening pictures, cross-referencing them for a quick match, extracting features from every image, locating anomalies with established rules, and providing recommendations to assist human operators in their decision-making. Multi-tasked with all these functions, an automatic recognition algorithm can serve as the super brain that assists human operators for image interpretation in a more timely and accurate manner.

As for now, ATR has transformed from concept to reality. Committed to combatting the illegal trade in cigarettes and alcohol, Customs authorities from Asia-Pacific and Europe are now applying the full-fledged ATR algorithms. As an add-on tool for an Asia-Pacific Customs service, ATR predetermines automatically if the scanning image contains any suspicious objects, while removing the necessity to analyse repetitive images of similar frequent goods.

In another three-month blind test by an Asia-Pacific Customs administration, ATR application has proved its capacity to detect cigarettes with a detection probability of almost 88%, and alcohol with an amazingly high accuracy rate of 93%. It should be noted that ATR application can achieve an even higher detection rate through further ‘machine learning’ in synchronization with the completed image database.

More information
chenzhiqiang@nuctech.com
www.nuctech.com