Artificial Intelligence Breakthrough in UXO Classification

Strategic Insights -

Research from the Silesian University of Technology, in collaboration with Polish institutions, has achieved a significant milestone in the development of artificial intelligence for Unexploded Ordnance (UXO) classification. By leveraging a high-fidelity digital twin and a custom Convolutional Neural Network (CNN), the team has demonstrated an 84.65% balanced accuracy rate, surpassing traditional models in the presence of distortions. This innovation holds considerable potential for real-time onboard classification in underwater vehicle missions. Furthermore, the study offers actionable suggestions for improving both model deployment and data acquisition protocols in the field.

Key Takeaways:

  • The classification of UXO from magnetometer data is a critical task, often hindered by data scarcity, and can be approached with a high-fidelity digital twin to generate comprehensive datasets.
  • The custom CNN model designed by the research team achieves a balanced accuracy of 84.65%, outperforming traditional models under distortions such as additive noise, drift, and time-wrapping.
  • A compact two-block CNN variant retains competitive accuracy while reducing the number of learnable parameters by approximately 33%, making it suitable for real-time onboard classification in underwater vehicle missions.
  • Architectural components such as residual skip connections and element-wise batch normalization are crucial for achieving model stability and performance.
  • The research provides actionable suggestions for improving both model deployment and data acquisition protocols in the field.
  • The study's findings emphasize the practical implications of underwater vehicles for survey design, highlighting the need to mitigate signal drift and maintain constant survey speeds.
  • The research was financially supported by Silesian University of Technology, Polish Ministry of Science And Higher Education, and Polish National Centre For Research And Development.

Statistics:

  • The custom CNN model achieves a balanced accuracy of 84.65%.
  • The compact two-block CNN variant reduces the number of learnable parameters by approximately 33%.
  • The research emphasizes the importance of maintaining constant survey speeds to mitigate signal drift.

Sources:

  • On the Development of a Neural Network Architecture for Magnetometer-Based UXO Classification. Applied Sciences, 2025,15(15):8274. (Applied Sciences - http://www.mdpi.com/journal/applsci)
  • NewsRx. Data from Silesian University of Technology Broaden Understanding of Machine Learning (On the Development of a Neural Network Architecture for Magnetometer-Based UXO Classification). Journal of Engineering. August 25, 2025; p 307.