Application and Advances in Radiographic and Novel Technologies Used for Non-Intrusive Object Inspection

Dmytro Mamchur, Janis Peksa, Soledad Le Clainche, Ricardo VinuesaSensors 2022 · DOI

TL;DR

This paper reviews current and emerging technologies for non-intrusive object inspection, particularly at border control points. It highlights the limitations of traditional X-ray methods and explores advancements in computed tomography, multi-energy imaging, and novel sensor technologies like electronic noses and biosensors, emphasizing the role of AI and machine learning in improving detection accuracy and efficiency.

What Problem Does This Paper Address?

The increasing volume of global trade and travel, while beneficial for economic growth and cultural exchange, also presents significant security challenges. Malicious actors exploit these open borders to smuggle illicit goods, weapons, and other dangerous materials. Traditional inspection methods, while effective for decades, are increasingly being outmaneuvered by sophisticated criminal tactics. This necessitates the development and implementation of advanced, non-intrusive inspection technologies that can efficiently and accurately identify threats without unduly delaying legitimate passengers and cargo. The paper addresses the need for a comprehensive overview of existing and developing technologies that can meet these evolving security demands.

Methods

The paper provides a comprehensive review of various non-intrusive inspection technologies. It categorizes these into:

  • Radiographic Techniques:
    • Planar Radiography: The most common technique, providing 2D images based on X-ray absorption. Limitations include superpositioning of objects and difficulty in determining depth and material composition.
  • X-ray Computed Tomography (CT): Offers 3D imaging by rotating X-ray sources and detectors, overcoming superposition issues and providing shape and depth information.
  • Dual- and Multi-Energy Imaging: Enhances material identification by analyzing X-ray attenuation at different energy levels, allowing for classification of materials into organic, non-organic, and mixtures.
  • Backscatter Techniques: Detects reflected X-rays, useful for identifying materials with different densities.
  • X-ray Diffraction Imaging (XRD): Analyzes the molecular structure of materials, effective for identifying specific substances but currently limited in scope.
  • Novel Technologies:
    • Electronic Noses (e-noses): Mimic the sense of smell using sensor arrays to detect volatile organic compounds, showing promise for identifying drugs and explosives based on air analysis.
  • Mimicking Animal Sense Receptors: Inspired by biological sensory systems (e.g., crocodile skin), these sensors aim to detect mechanical, thermal, and chemical substances.
  • Other Techniques: Includes ultrasonic sensors and spectrography-based methods for detecting specific materials.
  • Results Analysis (AI/ML):
    • Machine Learning (ML): Discusses supervised, unsupervised, and reinforcement learning for pattern recognition and classification.
  • Artificial Neural Networks (ANNs): Explains their layered structure and application in image and sequence recognition.
  • Deep Learning: Highlights its ability to learn complex patterns and features directly from data, particularly convolutional neural networks (CNNs) for image analysis and odor recognition.

Datasets

Not discussed in the paper. The paper focuses on a review of existing technologies and research trends rather than presenting new experimental results or datasets.

Key Insights and Ideas

  • The arms race in security: Criminals are constantly evolving their methods to bypass inspection, necessitating continuous innovation in detection technologies.
  • Beyond shape: Material matters: While X-ray imaging excels at revealing object shapes, accurately identifying material composition is crucial for distinguishing threats from benign items. Dual- and multi-energy techniques are key here.
  • Nature as an inspiration: Biomimicry, particularly in developing "electronic noses" that mimic animal olfaction, offers a promising avenue for detecting trace amounts of dangerous substances.
  • AI as the great equalizer: Machine learning and deep learning are transforming how inspection data is analyzed, enabling more sophisticated pattern recognition and classification than ever before.
  • Integration is key: The future of non-intrusive inspection likely lies in the synergistic combination of multiple sensor types and advanced AI analysis, rather than relying on a single technology.

Key Results

The paper does not present new experimental results or quantitative performance metrics for novel technologies. Instead, it synthesizes findings from existing research:

  • X-ray CT and multi-energy systems have demonstrated effectiveness in detecting various types of explosives and identifying material compositions.
  • Electronic noses have shown efficacy in detecting specific explosive substances like TNT and PETN by analyzing air samples.
  • AI-based pattern recognition and classification techniques, particularly deep learning, are reported to provide highly reliable recognition and classification results.
  • The paper highlights that while existing techniques are successful, there are still incidents of illicit trading, smuggling, and terrorism, indicating a need for further improvement.

Limitations and Future Work

  • Cost and Complexity: Advanced techniques like multi-energy CT are more expensive and require more complex hardware and software, limiting their widespread adoption.
  • Scope of Detection: Some techniques, like X-ray diffraction, are currently limited in the types of materials they can identify (e.g., liquid explosives remain a challenge).
  • Data Requirements for AI: Training robust AI models, especially deep learning models, requires large volumes of diverse datasets, which can be challenging to acquire and label.
  • Future Work: The authors suggest that improving non-intrusive inspection experience can be achieved by combining novel sensor types (like biosensors) with traditional techniques (like X-ray inspection). They also emphasize the potential of interpretable deep learning methods to significantly enhance the performance of these applications. Further research is needed to overcome limitations in current techniques and to develop more comprehensive and efficient inspection systems.

Summary updated

BibTeX
@article{mamchur2022application,
  author = {Dmytro Mamchur and Janis Peksa and Soledad Le Clainche and Ricardo Vinuesa},
  title = {{Application and Advances in Radiographic and Novel Technologies Used for Non-Intrusive Object Inspection}},
  journal = {Sensors},
  year = {2022},
  doi = {10.3390/s22062121},
}
Figure