By Murat Aydoğmuş, Lieutenant Commander (OF-3) TUR-N, Electronic Warfare Specialist, C-IED Centre of Excellence
Improvised Explosive Devices (IEDs) represent a persistent and evolving threat in various conflict zones and areas of instability, demanding continuous advancements in detection and neutralization techniques in all domains. This article provides an examination of the methods used to address this challenge, tracing the evolution from traditional approaches to contemporary deep learning (DL) solutions for countermeasures against IEDs like detection and classification technologies. Initially, some conventional techniques are reviewed, including visual inspection, metal detectors and chemical sniffer systems, highlighting their operational principles. Building on this foundation, modern advancements are explored, focusing on DL methodologies that have transformed IED detection and classification. Specifically, some state-of-the-art applications like Convolutional Neural Networks (CNNs) for enhanced imaging analysis, Object Detection frameworks such as YOLO, Faster R-CNN and SSD for real-time identification are described briefly in this article. Additionally, integration of these models into unmanned systems to improve the process is investigated, and AI-driven decision support systems that assist in strategic responses are examined as future advances. This article also addresses the challenges associated with these modern techniques, including data requirements, computational demands, and system integration issues. By examining some recent studies, case analyses, and practical applications, comprehensive understanding of the status of IED detection and classification is addressed. Additionally, future research directions aimed at improving the effectiveness and safety of these systems are provided. The findings underscore the transformative impact of DL on IED countermeasures and the ongoing need for innovation in this critical area.
INTRODUCTION
The detection, classification and neutralization of Improvised Explosive Devices (IEDs) remain critical challenges for security forces worldwide. IEDs pose a significant threat in both conflict zones and civilian environments, necessitating the development of effective countermeasures1 . Various approaches have been employed to mitigate this threat, ranging from conventional methods like visual inspection and metal detection to advanced deep learning (DL) models capable of autonomously identifying and classifying IEDs. This evolution reflects the need for a comprehensive understanding of technologies that can both detect, classify and neutralize IEDs in diverse operational contexts2.
Existing counter-IED (C-IED) strategies often lack integration between traditional and modern technological solutions, making it crucial to explore these methodologies collectively to enhance operational effectiveness3.
This survey article provides a systematic review of the state-of-the-art DL techniques in IED detection and classification, beginning with traditional techniques and progressing to advanced models. By examining the strengths and limitations of the approaches, this article aims to offer insights into how these technologies can be integrated into a comprehensive C-IED framework. The goal is to inform the community of interest and technology developers about the capabilities and gaps in current C-IED efforts.
This article aims to support the development of robust, integrated C-IED strategies that not only meet current operational needs but also anticipate future threats posed by increasingly sophisticated IED actors. Through this analysis, the paper seeks to advance the conversation on how to leverage traditional and emerging technologies to enhance global C-IED efforts.
CONVENTIONAL TECHNIQUES
Visual Inspection
Visual inspection aims to detect any signs of tampering, suspicious objects, or unusual disturbances that could indicate an IED. This method is often performed by highly trained personnel or robots. Inspecting for unusual items like wires, packages, or disturbances in soil. Looking for common IED concealment areas, such as vehicles, garbage bins, or under the ground. Signs like freshly dug ground, disturbed areas, or exposed wires might indicate the presence of an IED. It can be dangerous and time-consuming since the IED might not always be visible. Visual inspection also relies heavily on human expertise 4.
Metal Detectors
Metal detectors are used to locate metal components in IEDs, such as the casing or wiring. While many IEDs are designed to have minimal metal to avoid detection, metal detectors are still useful in many scenarios. The detector emits electromagnetic fields, and when it passes over metallic objects, the device detects a disturbance and sends a signal. These are commonly used in tandem with ground-penetrating radar (GPR) or other sensor technologies to detect IEDs beneath the surface. IEDs made of non-metallic materials can bypass detection 5 . Also, environments with a high amount of metal debris can lead to false positives5 ,6.
Chemical Sniffer Systems
Chemical sniffer systems are used in IED detection to identify explosive materials by detecting specific chemical signatures associated with explosives. These systems utilize various technologies, such as ion mobility spectrometry, mass spectrometry, and surface acoustic wave sensors, to analyze air samples for trace vapors emitted by explosives7,8,9. They can be deployed in handheld devices or integrated into larger systems, allowing for real-time detection in various environments. Chemical sniffer systems enhance the capability to locate IEDs by identifying low concentrations of explosive substances, thereby improving overall security and threat assessment.
Ion Mobility Spectrometry (IMS) detects chemicals based on the mobility of ions in an electric field. Explosive molecules ionize and travel at different speeds depending on their size and shape. Widely used in handheld chemical sniffers and airport security scanners7 .
Mass Spectrometry (MS) identifies chemical compounds based on the mass-to-charge ratio of ions. It’s highly sensitive and can detect trace amounts of explosive materials. More commonly used in laboratory settings but can be integrated into mobile devices8 .
Surface Acoustic Wave (SAW) These sensors detect changes in the frequency of surface acoustic waves when explosive molecules bind to a coating on the sensor surface. Employed in portable devices for real- time detection in the field9.
Laser-Based Detection
Laser-based IED detection involves using laser technology to identify explosive materials by analyzing their chemical signatures. Techniques such as Raman spectroscopy utilize laser light to detect the vibrational modes of molecules, helping to distinguish specific explosive compounds10. Quantum cascade lasers can measure how gases absorb light at wavelengths, indicating the presence of explosive vapors11. These methods offer high sensitivity and specificity for detecting explosives from a distance, making them valuable tools in enhancing the safety and efficiency of IED detection operations in various environments.
Canine Detection
Although not a mechanical system, trained dogs are highly effective at detecting explosives due to their acute sense of smell. They remain widely used in conjunction with chemical sensors12.
DEEP LEARNING BASED APPROACHES
On the other hand, with the developed technology both in terms of sensing materials and devices and their processing techniques propose a more automized chain of operations in IED detection and neutralization. Thanks to the state-of-the-art deep learning (DL) approaches, the process has been easier in recent years.
Convolutional Neural Networks (CNNs) have become increasingly popular for detecting IEDs due to their ability to learn complex patterns from visual data. There are different kinds of imaging techniques of the data of interest. In a basic DL approach of IED detection:
- Images are captured from various sources, including surveillance cameras, drones, or ground- based vehicles etc,
- Standard CNN architectures or custom-designed models are implemented to detect/classify images if they contain IEDs or not,
- Some Techniques such as rotation, scaling, and flipping of the image are applied to increase the number of data for training and validation, thus improving model robustness.

Figure 1: Simple illustration of classification as IED or not IED with a deep-learning (CNN) structure.
On the other hand, when we think of object detection frameworks like YOLO (You Only Look Once)13, Faster R-CNN14 and SSD (Single Shot MultiBox Detector)15, they provide many advantages and conveniences in the scope of automation. To give information very briefly about these methods:
- YOLO is a popular framework that can be used for real-time detection of different types of IEDs. YOLO can be trained to recognize different types of IEDs and their components.
- Another popular DL based IED detection method is Faster R-CNN. This method combines region proposal networks with CNNs to effectively identify and classify objects. This approach can be used to pinpoint the location of IEDs in images.
- SSD is another real-time object detection framework that can be used for IED detection in various scenarios.
These are all very common and popular tools to detect and classify IEDs and these DL based detection methods can be increased. They require a large amount of training datasets, but after a certain point, they can offer significant advantages to operators or users with their highly accurate decision- making abilities.
IEDs can take on any shape or form. Therefore, deep learning-based detection and classification require much more detail about the threat during the training phase. One of the studies shows that different shapes of IEDs, such as cylinders, squares, and spheres, can be successfully detected and classified by the deep learning algorithm based on their shapes. This means that if the database contains sufficiently rich data, it will be possible to classify IEDs and automatically alert the user using deep learning methods. This approach is applicable to systems that extract information about the target from the obtained images16.
One example on using x-ray images of luggage and containers that might contain IEDs: Images of the bags, containers are obtained with an x-ray system mounted on an autonomous vehicle and the presence of explosives is automatically reported to the users of the system or operators. Thus, the automatic detection of IEDs based on DL and artificial intelligence systems during the execution of critical tasks offers significant advantages in terms of both cost and minimizing human casualties17.
ADVANTAGES OF THE DL BASED MODELS
DL models are excellent at extracting detailed features from sensor data, which leads to more accurate identification of IEDs. Additionally, DL based models can recognize complex patterns in the data that might be missed by traditional methods, resulting in better detection rates.
Deep learning models can be trained on varied datasets, allowing them to recognize the features of different IED types in various settings, such as urban or rural areas. These models are flexible and can adapt to different scenarios. Additionally, pre-trained models can be easily adjusted to detect new explosives or operate in different environments with minimal extra data, speeding up deployment and enhancing their capacity to handle new conditions.
After training, DL models can rapidly analyze large amounts of sensor data, enabling real-time detection of IEDs. Furthermore, DL can be incorporated into autonomous systems such as drones or robots, facilitating continuous and automated monitoring without the need for ongoing human oversight. DL models help reduce the number of false positives, which is important for avoiding unnecessary alerts or interruptions during operations.
Another important advantage of DL models is that they perform well even when dealing with noisy data. They are more robust in complex environments, such as crowded areas or rough terrains, where IEDs might be hidden.
Automating IED detection using deep learning decreases reliance on manual searches or continuous human supervision, which are often time-intensive and costly. These models can be swiftly implemented in operational environments without the need for significant extra resources, making them more economical over time.
LIMITATIONS OF DL BASED MODELS
Besides their advantages, DL systems have limitations when it comes to disarming or neutralizing IEDs. At least for now, they cannot operate independently without human supervision because the complex nature of real-time decision-making in these situations often requires human intervention for safety.
Additionally, the effectiveness of deep learning models depends on the availability of large datasets for accurate detection and classification of IEDs. However, gathering such datasets can be challenging due to the diverse range of IED designs and the various environments in which they may be encountered.
In summary, DL based IED detection systems, which offer significant advantages, especially in terms of automation, still contain some limitations and areas that need further development.
WHAT TO EXPECT IN THE FUTURE?
The future of using DL methods in detecting IEDs has promising potential for significant advancements in various areas. One direction could focus on enhancing the integration of data from multiple sensors (e.g. optical, infrared, x-ray, acoustic, chemical) using advanced DL models. Thanks to the AI techniques, this multi-modal data can be combined to provide more accurate and real-time detection, reducing false positives and improving IED identification under diverse weather and terrain conditions. Moreover, future models may become more context-aware by using DL to combine environmental factors, human activity data, and sensor input to assess the likelihood of IED presence, improving precision in complex scenarios.
By incorporating DL in multi-agent systems, where robots or drones work together with human teams, can create seamless collaboration. These systems can integrate real-time sensor data and interact dynamically to detect and neutralize IEDs in complex, hazardous environments.
One of the very important directions is to adapt to novel IED designs and emplacement strategies more quickly. Thanks to the DL based models, new types of IEDs with minimal data could be generalized and detection and classification probabilities can be increased.
Since knowledge sharing or namely transfer learning across different domains is very important for AI systems, new research could explore how DL techniques developed for detecting IEDs in military environments can be transferred to other domains, such as border security, airport screening, or critical infrastructure protection. Cross-domain knowledge transfer can enhance the applicability of these technologies across a broader range of threat detectionscenarios.
In conclusion, the future work of AI and deep learning in IED detection will likely involve more advanced, adaptive, and real-time systems that leverage multiple data sources, work autonomously and collaboratively, and integrate seamlessly with human decision-making. With continuous innovations in neural networks, robotics and sensor technologies, the potential for AI to improve IED detection and classification will expand further in the coming years. ■
REFERENCES
- United Nations, “CTED Guidelines: Enhancing the Capacity of Member States to Prevent and Combat Terrorism,” UN Counter-Terrorism Committee Executive Directorate, Jan. 25, 2023. [Online]. Available: https://www.un.org/ counterterrorism/sites/www.un.org.counterterrorism/files/ 20230125_cted_guidelines_rev_en.pdf. [Accessed: Sep. 03, 2024].
- United Nations Institute for Disarmament Research (UNIDIR), “Explosive Weapons in Populated Areas: A Guide for States,” UNIDIR, 2023. [Online]. Available: https://unidir.org/fi les/publication/pdfs/-en-641.pdf. [Accessed: Sep. 05, 2024].
- North Atlantic Treaty Organization (NATO), “Countering Improvised Explosive Devices,” NATO, [Online]. Available: https://www.nato.int/cps/en/natohq/topics_77646.htm. [Accessed: Sep. 05, 2024].
- United States Marine Corps, “Improvised Explosive Device (IED) Awareness,” Training and Education Command, [Online]. Available: https://www.trngcmd.marines.mil/Portals/ 207/Docs/TBS/B3L0487XQ-DM%20Improvised%20 Explosive%20Device.pdf?ve. [Accessed: Sep. 07, 2024].
- L. E. Decker, J. H. Lentz, and R. A. Miele, “The impact of improvised explosive devices on military operations”. *Naval Submarine Medical Research Laboratory*, vol. 28, no. 2, pp. 234-240, 2022. [Online]. Available: https://onlinelibrary.wiley.com/doi/10.1002/nsg.12282. [Accessed: Sep. 06, 2024].
- Wikipedia, “Metal Detector,” Wikipedia, The Free Encyclopedia, [Online]. Available: https://en.wikipedia.org/ wiki/Metal_detector. [Accessed: Sep. 06, 2024].
- Fetterolf, D.D. (1993). Detection of Trace Explosive Evidence by Ion Mobility Spectrometry. In: Yinon, J. (eds) Advances in Analysis and Detection of Explosives. Springer, Dordrecht. https://doi.org/10.1007/978-94-017-0639-1_13
- J. Yinon, “Analysis and Detection of Explosives by Mass Spectrometry,” in Aspects of Explosives Detection, M. Marshall and J. C. Oxley, Eds. Elsevier, 2009, pp. 147–169. doi: 10.1016/B978-0-12-374533-0.00008-8.
- A. Mujahid and F. L. Dickert, “Surface Acoustic Wave (SAW) for Chemical Sensing Applications of Recognition Layers,” Sensors, vol. 17, no. 12, p. 2716, 2017. doi: 10.3390/s17122716.
- M. E. Farrell, E. L. Holthoff, and P. M. Pellegrino, ‘Laser- based optical detection of explosives’, 01 2015, pp. 1–8.
- Block Engineering, “Active Detection of Explosives: A Technology Overview,” Block Engineering, [Online]. Available: https://www.blockeng.com/technology/active.pdf. [Accessed: Sep. 16, 2024].
- K. Frank, H. Holness, K. Furton, and L. DeGreeff, ‘Explosives detection by dogs’, 01 2022, pp. 47–75.
- V7 Labs, “YOLO Object Detection: An Overview,” V7 Labs, [Online]. Available: https://www.v7labs.com/blog/ yolo-object-detection. [Accessed: Sep. 02, 2024].
- R. Galvez and E. Dadios, ‘Identifying threat objects using faster region-based convolutional neural networks (faster R-CNN)’, International Journal of Advances in Intelligent Informatics, vol. 8, p. 381, 11 2022.
- W. Liu, D. Anguelov, D. Erhan, C. Szegedy, S. Reed, C.-Y. Fu, and A. C. Berg, “SSD: Single Shot MultiBox Detector,” in *Computer Vision – ECCV 2016*, Springer International Publishing, 2016, pp. 21–37. doi: 10.1007/978-3-319-46448- 0_2.
- S. Lameri, F. Lombardi, P. Bestagini, M. Lualdi, and S. Tubaro, “Landmine detection from GPR data using convolutional neural networks,” in *25th European Signal Processing Conference, EUSIPCO 2017*, vol. 2017, 2017, pp. 508–512. doi: 10.23919/EUSIPCO.2017.8081259.
- C. Chamnanphan, S. Vorapatratorn, K. Kirimasthong, T. Boongoen, and N. Iam-On, ‘Improvised Explosive Device Detection Using CNN With X-Ray Images’, Journal of Advances in Information Technology, vol. 14, pp. 674–684, 01 2023.
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 masters’ degrees in Electrical Engineering and Engineering Acoustics from the Naval Postgraduate School. His background is primarily based on Navy R&D projects. He is currently pursuing PhD studies in underwater target detection using modern deep-learning models.
Email: maydogmus@ciedcoe.org
Counter-IED Report Autumn 2024