Improving Underwater Capabilities: The impact of Artificial Intelligence on IED detection and identification

 

By Murat Aydoğmuş, Lieutenant Commander (OF-3) TUR-N, Electronic Warfare Specialist, C-IED Centre of Excellence

INTRODUCTION

Detecting and identifying Improvised Explosive Devices (IEDs) in underwater environments presents significant operational challenges. Unlike traditional explosives, IEDs are designed in an improvised way and often activated remotely, which makes them a continuous threat in both military and civilian maritime contexts. With the increasing importance and reliance on naval and offshore infrastructure for trade, energy production, and defense, the risk posed by the underwater threat has become a critical concern for global security. More recent incidents, such as the detonation of limpet mines illustrate the significant risks posed by explosives that can be manufactured with relative ease and evolve into asymmetric threats1.

Detecting IEDs and other potentially suspicious underwater objects represents a critical component of operations. These threats present significant risks not only to naval and commercial assets but also to human life. Ensuring the safe identification and handling of such underwater hazards is therefore of great importance. Traditional detection methods, such as sonar scanning and manual diver inspections, have proven time-consuming, labor-intensive, and risky, particularly in deep-sea environments or areas with poor visibility. Advancements in artificial intelligence (AI) have revolutionized the field of underwater threat detection. Tools developed with AI support have improved the precision and effectiveness of detecting underwater objects2. Deep learning models trained on extensive sonar and optical image datasets can quickly identify possible threats, minimize false alarms, and enhance overall situational awareness. Additionally, Autonomous Underwater Vehicles (AUVs) equipped with AI-driven navigation and object recognition capabilities can autonomously patrol high-risk areas, minimizing the need for human divers and enhancing operational safety. Research efforts around the world have focused on developing AI-supported sonar processing to improve detection capabilities in complex marine environments3.

As underwater threats continue to evolve, the integration of AI solutions remains an important aspect of modern naval defense and maritime security. This article examines the technological advancements in AI- aided detection and identification of small underwater targets or threats with an emphasis on IEDs, highlighting the benefits, challenges, and future potential of these systems3.

TRADITIONAL METHODS

To detect and classify underwater IEDs or potential suspicious objects, there are some traditional approaches like using sonar scanning and analysing the screen, magnetic anomaly detection (MAD), diver-based inspections, and remotely operated vehicles (ROVs). The primary technology used for underwater detection is sonar systems. Sonar devices rely on sound waves to detect objects or threats underwater because electromagnetic waves are heavily absorbed and attenuated by water, making them ineffective for communication or detection.

Sonar devices have different monitoring technology according to the system’s active or passive operation mode. While passive systems are used to listen to the environment without any emissions, active systems send out pulses and receive the reflections. In this article, examples will be given from active and high- resolution sonar systems. Although there are different kinds of sonar systems used for underwater monitoring like multi-beam echo sounders, forward-looking sonar, side-scan sonar (SSS), and synthetic aperture sonar (SAS) systems, the use of SSS and SAS operation will be briefly explained in this article. Both sonar systems provide detailed images of the underwater environment.

Magnetic anomaly detection (MAD), on the other hand, identifies changes in the Earth’s magnetic field caused by metallic objects, and these systems are used traditionally to detect anomalies and potential underwater targets4. While MAD is particularly useful for detecting metal-based threats, it is less effective when it comes to non-metallic IEDs.

Sonar System Operations:

The underwater environment poses many challenges because of the complex characteristics of water. Several sonar technologies are used for seabed scanning and small object detection, and these systems typically operate at higher frequencies, which allows for high-resolution imaging.

In the traditional approach, detecting small objects relies heavily on manual analysis of sonar images, a process that is time-consuming and highly dependent on the skill of the operator. This dependence can lead to inconsistent outcomes and limits scalability, particularly when large amounts of data need to be processed. The inherent complexity of the underwater environment adds an additional layer of difficulty to detection operations5. Factors such as noise, clutter, and varying seabed conditions make it difficult for operators to accurately identify objects manually.

a. Side Scan Sonar (SSS):

SSS is an important tool for detecting underwater objects, including IEDs and other small objects on the seabed. SSS works by emitting conical or fan-shaped pulses that are directed downward toward the seabed at a wide angle, perpendicular to the sensor’s path through the water. The sensor can be towed behind a surface vessel or submarine, or it can be mounted on the ship’s or unmanned vehicle’s hull for more stationary scanning.

The sonar emits sound waves, which interact with the seafloor and any objects resting on or buried within it. These sound waves reflect to the sonar system, and the intensity of these acoustic reflections is recorded. The resulting data is captured in a series of cross-track slices, which can then be analyzed to reveal anomalies on the seafloor.

This sonar is useful for detecting metallic objects like underwater mines, IEDs, and debris, as these typically produce strong acoustic reflections. However, side-scan sonar can also detect non-metallic objects when their shape and density exhibit significant contrast relative to the surrounding environment. For example, irregularities caused by buried objects or disruptions in the seafloor structure might indicate the presence of IEDs, which are often buried or camouflaged6.

Figure1: Illustration of SSS Operation.7

b. Synthetic Aperture Sonar (SAS):

SAS is also a powerful tool for obtaining high-resolution underwater images and they can also be used for detecting the objects like mines, underwater IEDs or other potential objects. SAS, which is mounted on a UUV, moves along a track and synthesizes a much larger virtual array from many pings. It continuously records backscattered signals from the seafloor or the objects. The system coherently combines the echoes received at different positions, effectively synthesizing a large aperture. Much finer resolution is obtained with SAS8.

Regarding the detection of IEDs or small buried objects underwater; they offer more detailed imagery of the seafloor and large areas are scanned efficiently with high-resolution outputs.

Figure 2: Illustration of SAS Operation.8

Magnetic Anomaly Detection (MAD) Operations:

MAD systems utilize sensitive magnetometers designed to detect perturbations in the Earth’s magnetic field induced by the presence of ferrous objects9. Normally, the Earth’s magnetic field is quite stable, and natural underwater materials like sand or water don’t affect it much. But when there is a metal object – like a mine casing or parts of an IED – it might create a small disturbance in the magnetic field.

MAD sensors are used to detect these disturbances, known as magnetic anomalies. MAD sensors can be attached to ships, underwater drones, or even aircraft flying low over shallow waters. As the sensor moves, it constantly measures the magnetic field. If it detects an unusual magnetic signal, it marks that location as a potential threat. This is especially useful when IEDs are made with metal parts, and MAD can help find them even if they are buried under the seabed or not visible on sonar images. That’s one of the biggest strengths of MAD – it can detect hidden objects in cluttered underwater environments.

However, while MAD is great for detecting where a metallic object might be, it doesn’t tell us what the object is. A rock anchor, an old pipe, or a mine might all produce similar magnetic signals. So, once MAD finds something suspicious, other tools like sonar, cameras, or divers are usually needed to take a closer look and identify the object. The greatest limitations are that small devices have a weaker magnetic footprint, MAD only works effectively at close range, and small anomalies are only noticeable when background magnetic noise is low.

AI SUPPORT IN UNDERWATER MONITORING

In the traditional approach, operators visually inspect the high-resolution sonar imagery, and they look for shapes, shadows, textures, and other cues to identify potential threats like IEDs or mines, which is very dependent on experience and familiarity with underwater environments.

In the field of underwater target detection and identification, AI offers a transformative alternative to traditional methods by enabling automated processes that reduce or minimize human intervention. Unlike traditional approaches – which often rely on manual analysis, and an operator-dependent decision-making– AI-based systems utilize data-driven models to autonomously detect and identify underwater objects. This shift is particularly significant when considering the complex and dynamic nature of the underwater environment, where factors such as light absorption, turbidity, background clutter, and noise substantially hinder image clarity and object visibility.

The effectiveness of AI in such scenarios is heavily dependent on the quality of the input data. Obtaining high-resolution, noise-free imagery is essential for accurate detection and reliable identification. However, due to environmental constraints, such ideal conditions are not always achievable. Therefore, during the algorithm development phase, preprocessing tech-niques such as denoising, contrast enhancement, and super-resolution play a critical role in preparing the data for subsequent analysis. These preprocessing steps not only improve the visual quality of the images but also enhance the performance of AI models by ensuring that essential features are preserved and emphasized.

In addition, the variety and richness of a dataset have a direct impact on how robust and generalizable the trained models are. When a dataset covers diverse object appearances, angles, environmental scenarios, and sensor types, it allows AI systems to learn more detailed and representative features. As a result, this enhances detection performance and lowers the chances of false alarms or missed detections.

In recent years, there has been increasing interest in utilizing deep learning approaches – like convolutional neural networks (CNNs) and transformer-based models – for automatically detecting and classifying objects in underwater environments. These techniques have demonstrated strong potential in recognizing patterns and irregularities that might be challenging for human observers to detect. Even though IEDs are typically non-standard in shape and composition since they are constructed using improvised and unpredictable methods, AI-based approaches can still provide valuable insights. While they may not always deliver definitive identification, they are highly effective in flagging potential threats and drawing the operator’s attention to regions of interest.

Ultimately, the integration of AI into underwater detection systems enhances situational awareness, operational efficiency, and decision-making capabilities. By combining high-quality data acquisition, advanced preprocessing techniques, and powerful AI algorithms, it becomes possible to develop robust systems capable of performing in real-time, even under challenging conditions. This not only increases the safety of underwater operations but also reduces the cognitive load on human operators, enabling more focused and informed assessments of suspicious objects.

Although this article does not aim to explore the technical background in depth, it is important to highlight that significant research has been dedicated to the automatic detection of underwater objects. These studies span a broad methodological spectrum, encompassing single-stage detection models, two- stage detection frameworks, and more sophisticated multi-stage systems.

Single-stage methods integrate object localization and classification into a unified pipeline, enabling rapid inference and making them particularly advantageous for real-time applications and resource-constrained, modular systems. Conversely, two-stage and multi- stage approaches introduce intermediate processing steps – such as region proposal, candidate filtering, and refined classification – that, while computationally more demanding, often yield higher detection precision and robustness, especially in complex or visually cluttered underwater environments10.

Various algorithms have been developed within each of these methodological categories, and the research community continues to explore novel techniques and architectures in pursuit of improved performance. These efforts reflect the intrinsic challenges posed by the underwater domain, where visibility limitations, sensor noise, and the irregular, non-standardized construction of IEDs complicate reliable identification. As such, there is no single superior approach; rather, each method offers context- specific advantages, and their relative effectiveness is often determined by operational constraints such as processing time, hardware capabilities, and mission-critical requirements.

In this context, the ongoing development of AI-driven detection systems is of particular significance. Even when definitive classification of an object as an IED is not immediately possible, the ability to automatically flag anomalous or suspicious items contributes substantially to enhancing operator situational awareness and facilitating more focused human analysis. Therefore, the quality, speed, and adaptability of the detection algorithms play a crucial role in the overall efficacy of underwater surveillance and security operations.

Among more recent approaches, the integration of advanced payload systems on unmanned underwater vehicles (UUVs) has significantly enhanced target detection capabilities. These platforms are increasingly equipped with both optical and sonar imaging technologies11, enabling the collection of comple-mentary data sets. By applying multi-modal data fusion techniques – where optical imagery provides high-resolution detail and sonar imaging offers greater penetration in turbid or low-visibility environments, the overall accuracy and reliability of underwater detection systems can be substantially improved.

This synergistic use of sensor modalities allows for a more robust assessment of submerged objects, particularly in complex and dynamic underwater conditions. As a result, such technological advancements contribute to safer and more effective operations, whether in military reconnaissance or counter-IED missions.

As an example, limpet mines used as underwater IEDs are typically rudimentary and constructed with readily available materials, making them difficult to detect or defend against. The improvised nature of these devices means that they may lack the sophisticated triggering mechanisms of standard military mines, but this also makes it harder to trace and neutralize12. These mines may be manually placed by divers or deployed using UUVs, making their detection and identification a challenge. The use of AI-driven systems to detect and identify such underwater IEDs, including limpet mines, by utilizing advanced sonar imaging and optical technologies is a very good application area. As it was stated, when fused together, these techniques offer greater accuracy in distinguishing between benign objects and potential threats, contributing to safer operations in naval environments.

CONCLUSION

In recent years, there have been many research topics concentrated on leveraging AI-supported systems for underwater object detection. The primary goal of these studies is to improve situational awareness and ensure safer operations by minimizing the need for human intervention in risky underwater environments. The effectiveness of these automated detection systems heavily depends on the volume and diversity of the training data, as richer and higher-quality datasets allow AI models to perform more reliably across different environments and object variations.

Although the improvised nature of IEDs – characterized by their unpredictable shapes, materials, and deployment methods – poses significant challenges for reliable identification, the application of AI-supported tools plays a critical role in drawing attention to potential underwater threats. Ultimately, minimizing human exposure while maximizing detection reliability through automated systems not only improves safety but also strengthens the overall effectiveness of underwater threat mitigation strategies.■

REFERENCES
  1. The Maritime Executive, “Ambrey: Tanker at Ust-Luga May Have Been Hit With Limpet Mines,” The Maritime Executive, Feb. 12, 2025. [Online]. Available: https://maritime-executive.com/article/ambrey-tanker-at-ust-luga-may-have-been-hit-with-limpet-mines
  2. M. Kanguya, “Smart Oceans: The Rise of AI in Deep-Sea Exploration,” The Future List, Feb. 2025. [Online]. Available: https://www.thefuturelist.com/smart-oceans-the-rise-of-ai-in-deep-sea-exploration/
  3. M. Aubard, A. Madureira, L. Teixeira, and J. Pinto, “Sonar-based Deep Learning in Underwater Robotics: Overview, Robustness and Challenges,” arXiv preprint arXiv:2412.11840, 2024. [Online]. Available: https://arxiv.org/abs/2412.11840
  4. Clem, T.; Allen, G.; Bono, J.; McDonald, R.J.; Overway, D.; Sulzberge, G.; Kumarb, S.; King, D. Magnetic sensors for buried minehunting from small unmanned underwater vehicles. In Proceedings of the Oceans’04 MTS/IEEE Techno-Ocean ’04 (IEEE Cat. No.04CH37600), Kobe, Japan, 9–12 November 2004; Volume 2, pp. 902–910.
  5. F. Zhao, B. Huang, J. Wang, X. Shao, Q. Wu, D. Xi, Y. Liu, Y. Chen, G. Zhang, Z. Ren, J. Chen, and K. Mizuno, “Seafloor debris detection using underwater images and deep learning-driven image restoration: A case study from Koh Tao, Thailand,” Marine Pollution Bulletin, vol. 214, 117710, 2025. [Online]. Available: https://doi.org/10.1016/j.marpolbul.2025.117710
  6. NOAA Ocean Exploration, “Exploration Tools: Side-Scan Sonar,” NOAA Ocean Explorer. [Online]. Available: https://oceanexplorer.noaa.gov/technology/sonar/side-scan.html [Accessed: Mar. 24, 2025].
  7. Robert Wilson Marine Technologies (RWMT), “Side-Scan Sonar,” RWMT Surveying Tools. [Online]. Available: https://www.rwmt.se/surveying-tools/side-scan-sonar. [Accessed: Mar. 24, 2025].
  8. R. Edgar, “Introduction to Synthetic Aperture Sonar,” in Sonar Systems, InTechOpen, 2011. [Online]. Available: https://cdn.intechopen.com/pdfs/18868/InTech-Introduction_to_synthetic_aperture_sonar.pdf
  9. CAE, “Magnetic Anomaly Detection Extended Role (MAD-XR),” CAE, Mar. 2023. [Online]. Available: https://www.cae.com/media/media-center/documents/ datasheet.MAD-XR.pdf
  10. SharkYun, “Computer Vision — Object Detection, One-Stage vs Two-Stage,” Medium, Oct. 31, 2024. [Online]. Available: https://sharkyun.medium.com/computer-vision-object-detection-one-stage-vs-two-stage-b05dbff88195
  11. Kim, H.-G.; Seo, J.; Kim, S.M. Underwater Optical-Sonar Image Fusion Systems. Sensors 2022, 22, 8445. https://doi.org/10.3390/s22218445.
  12. J. D. Caverley, “From the Editors,” Naval War College Review, vol. 77, no. 3, pp. 1–2, Summer/Autumn 2024. [Online]. Available: https://digital-commons.usnwc.edu/cgi/viewcontent.cgi?article=1756&context=nwc-review
ABOUT THE AUTHOR

Lieutenant Commander Murat Aydoğmuş was commissioned into the C-IED Centre of Excellence as a C2 Electronic Warfare Specialist in the Defeat the Device Branch in August 2023. After assuming this position, he organized the 6th NATO C-IED Technology Workshop in Seville, Spain, in 2024. He holds master’s degree in Electrical Engineering and Engineering Acoustics from the Naval Postgraduate School. His background is primarily based on maritime projects and research and development activities in the maritime domain. He is currently pursuing PhD studies in underwater target detection using modern deep-learning models.

Email: [email protected]

Download PDF: 51-56 Murat Aydogmus article – IMPROVING UNDERWATER CAPABILITIES – COUNTER-IED REPORT, Spring-Summer 2025

COUNTER-IED REPORT, Spring/Summer 2025