Optimization of Deep Learning Parameters for Magneto-Impedance Sensor in Metal Detection and Classification

Area of Science:

• Sensor Technology
• Artificial Intelligence
• Signal Processing

Background:

• Deep learning excels with periodic data (e.g., electromyography, acoustic signals).
• Traditional deep learning models struggle with anomalous and irregular data from magneto-impedance (MI) sensors.
• MI sensors offer non-contact data acquisition, valuable for various applications.

Purpose of the Study:

• To develop and analyze a deep learning model optimized for MI sensor data.
• To enhance the detection and classification accuracy of irregular signals.
• To adapt deep learning for non-contact sensing applications.

Main Methods:

• Utilized a recurrent neural network (RNN) architecture combining Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) models.
• Configured and tested various RNN layers to optimize performance for MI sensor data.
• Implemented sequence length processing and refined prediction steps to improve accuracy.

Main Results:

• Achieved increased accuracy in detecting and classifying irregular MI sensor data compared to standard methods.
• Demonstrated the model's effectiveness in handling diverse and anomalous signal patterns.
• Validated the potential for improved performance through sequence length optimization and prediction refinement.

Conclusions:

• The proposed deep learning approach, integrating LSTM and GRU, is effective for analyzing irregular MI sensor data.
• This method significantly enhances the detection and classification of metal objects using MI sensors.
• The technology holds promise for diverse applications including drone control, autonomous driving, and foreign object detection.