Study on the Classification of Metal Objects by a Fluxgate Magnetometer Cube Structure.

Area of Science:

• Geophysics
• Machine Learning
• Robotics

Background:

• Unexploded bombs pose significant safety risks after conflicts.
• Accurate identification of underground metallic targets is critical for safe clearance operations.

Purpose of the Study:

• To develop and evaluate a deep learning system for classifying metallic object types using magnetic field data.
• To improve the accuracy of underground target recognition compared to traditional methods.

Main Methods:

• Constructed a fluxgate magnetometer cube arrangement structure (FMCAS) with an eight-sensor array.
• Acquired magnetic field data and created a diverse database of magnetic target models.
• Utilized an enhanced ResNet-18 deep learning network for classification.

Main Results:

• Achieved an 84.1% recognition accuracy on a validation set of 107 groups.
• Demonstrated high performance for targets with large magnetic moments, reaching 96.3% accuracy.
• Reported recall and precision rates of 96.4% for the best-performing targets.

Conclusions:

• The enhanced ResNet-18 network effectively classifies underground metallic items.
• This approach offers a novel method for underground metal target identification and classification.
• The system shows promise for enhancing safety in post-conflict zones.