Subscribe
US intelligence reports Iranian deployment of naval mines in Strait of Hormuz
Mines could threaten shipping alongside missiles and drones
Modern mines use magnetic, acoustic and pressure sensors to target vessels
Detection and clearance remain slow, complex and resource-intensive
AI and machine learning are increasingly used to identify mines via sonar
United States intelligence officials have assessed that Iranian forces have deployed a small number of mines in the Strait of Hormuz, a critical choke point for global shipping, according to reports. The move gives the Iranians a means, along with missiles and drones, of threatening ships.
The US Navy recently decommissioned the minesweeping vessels that it had operating in the Persian Gulf region. However, it has other ships and aircraft for finding and destroying mines.
As a computer scientist who researches how to detect mines, I have been researching how artificial intelligence techniques, such as machine learning, can help navies detect modern sea mines. Here’s what I’ve learned about how the mines work and how they can be neutralised.
The mines most people picture, like those seen in films such as “Godzilla Minus One,” are floating spheres tethered to the seabed, with small protrusions called Hertz horns that trigger the mine when it makes contact with a ship. These are called moored mines.
In the film, characters use a small wooden boat to sweep mines without triggering them because the mines responded to a metal-hulled ship’s magnetic field. Detecting magnetic fields is characteristic of influence mines, which respond to a ship’s magnetic, acoustic or pressure signature, as opposed to simple contact mines that detonate when ships run into them.
Modern mines typically combine multiple sensing modes. Some are designed to detonate only after a certain number of ships have passed, allowing them to ignore smaller vessels or minesweeping attempts and target higher-value ships. Examples include the Iranian Maham 3, which uses both magnetic and acoustic sensors.
Not all mines float. Many modern mines instead sit on the seabed. These mines are most effective in shallow water, where ships pass closer to the seabed. Some bottom mines sit exposed on the seabed, while others are partially or completely buried in sediment. Examples include the Iranian Maham 7 and the Manta mine, a low-profile bottom mine used by Iraq during the 1991 Gulf War. These mines can be deployed by small vessels or laid from aircraft, making them relatively easy to place. They are triggered when they sense a ship passing overhead.
Many modern mines are cylindrical or torpedo-shaped, allowing them to be deployed from aircraft or submarines and descend in a controlled way before settling on the seabed. More advanced designs include so-called rising mines, which sit on the seabed and launch upward toward a target once it is detected.
A key advantage of naval mines is not just the damage they can cause, but also the time and resources required to find and clear them. This is because it’s challenging to do so over large areas quickly and reliably.
Even the possibility of mines can disrupt shipping and force extensive and costly clearance operations. This has been demonstrated in practice: During the 1980s, Iran and Iraq deployed relatively small numbers of mines against each other in the so-called Tanker War in the Persian Gulf and Red Sea. This caused significant disruption to shipping and forced costly, time-consuming clearance operations, even when direct damage was limited.
Some countermeasures use uncrewed systems to trigger mines by mimicking the magnetic or acoustic signatures of ships, or to disable them with explosive charges. However, more targeted approaches require identifying individual mines, which motivates the need for reliable detection.
Mine detection is best understood as a wide-area sonar search, which produces many contacts — essentially, anything unusual in the sonar data. Automatic target recognition algorithms then triage these contacts and classify them as either minelike objects or benign. Divers or camera systems then provide higher-confidence identification or confirmation to validate the result. This is known as a detect-classify-identify pipeline.
To collect data, an uncrewed surface vehicle — deployed from a larger ship — can tow a sonar platform at a fixed height above the seabed. The platform, called a towfish, resembles a small missile and carries multiple sensors, including port and starboard side-scan sonar. The British Royal Navy is also preparing to send this type of towed sonar array to the Persian Gulf region, according to a report.
These sonar devices use sound rather than light to form images. Unlike a photograph, a sonar image is built from one-dimensional measurements of returned sound energy as a function of distance from the sensor. As the platform moves, these slices are assembled to form a continuous image of the seabed. The center of the image corresponds to the water column directly beneath the sonar device and appears dark. The seabed appears as if illuminated from the sensor, with objects characterized by a bright highlight facing the sonar and a shadow extending away from it.
At the detection stage, researchers have developed a range of techniques to detect minelike objects in sonar imagery. Early methods segmented sonar imagery into regions that show as highlights paired with acoustic shadows. Other statistical approaches model seabeds and identify anomalies that deviate from it. Template-like matched filters are used to identify objects with known geometric characteristics.
More advanced approaches incorporate machine learning, using carefully selected features derived from texture, intensity and shadow geometry to classify objects.
More recently, researchers have applied deep learning methods directly to sonar imagery and have often shown improved performance, particularly in complex environments. But their effectiveness depends on the availability of representative training data.
Unlike the data for training many other computer vision systems, high-resolution side-scanning sonar data is particularly expensive to collect and label in large enough amounts to successfully train deep learning mine detection systems.
Perhaps, when it becomes safe to do so, navies can clear mines from the Strait of Hormuz and add to the limited supply of this data.
John Femiani, Associate Professor of Computer Science and Software Engineering, Miami University
This article is republished from The Conversation under a Creative Commons license. Read the original article.