Neural Network for Metal Detection Based on Magnetic Impedance Sensor.

Area of Science:

• Sensor Technology
• Artificial Intelligence
• Electromagnetism

Background:

• Magnetic impedance (MI) sensors detect metal objects by sensing magnetic field changes.
• Detecting metal objects with MI sensors is challenging due to small, noisy magnetic field variations, limiting detection distance.
• Deep learning offers a potential solution to improve the sensitivity and range of MI-based metal detection.

Purpose of the Study:

• To investigate the efficiency of deep learning methods for metal detection using data from multiple MI sensors.
• To compare the performance of Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) in analyzing MI sensor data for metal detection.
• To analyze the performance of a deep-learning-based (DLB) metal detection network incorporating multiple MI sensors, Long Short-Term Memory (LSTM), and CNNs.

Main Methods:

• Utilized data from multiple magnetic impedance (MI) sensors for metal detection.
• Applied deep learning models, specifically Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs), to analyze sensor data.
• Compared the performance of CNN and RNN models, including variations in network depth and metal sheet size, using LSTM and CNN integration.

Main Results:

• Recurrent Neural Networks (RNNs) demonstrated superior overall performance compared to Convolutional Neural Networks (CNNs) for metal detection using MI sensor data.
• Convolutional Neural Networks (CNNs) showed better performance in the initial stages of detection compared to RNNs.
• The deep-learning-based (DLB) network performance was analyzed based on the number of network layers and the size of the detected metal sheet.

Conclusions:

• Deep learning, particularly RNNs, significantly improves metal detection capabilities with multiple MI sensors.
• The study highlights the potential of combining different deep learning architectures like LSTM and CNN for enhanced sensor-based detection.
• Findings are expected to advance sensor-based deep learning detection technologies for improved metal object identification.