Fast Real-Time Electromagnetic Tomography System for Size Prediction and Material and Position Classification

Electromagnetic tomography (EMT) finds extensive applications in characterizing the property distribution of conductive/magnetic materials. However, real-time online target classification and recognition using EMT have not been investigated, which provides significance for automatic detection in the industry domain. This article presents a miniature EMT sensor designed for real-time monitoring and classification of small targets. The EMT system integrates a real-time algorithm with deep learning (DL) techniques to enhance accuracy and efficiency in target detection and classification. First, image reconstruction is carried out with different inverse algorithms on simulated and measured data. The performance of the algorithms has a consistent rank for different target measurements under different combinations with size, material, and position, where the optimized single value decomposition (SVD) algorithm contributes to the highest image quality and good response. In addition, images reconstructed using the optimized SVD algorithm are input into the proposed lightweight simplified ResNet network with a self-attention mechanism, enabling real-time size prediction as well as material and position classification. In the simplified network, the response time of prediction is 0.0375 s for each image, which reaches the requirement of real-time monitoring. The average errors of estimated sizes are 0.0158 mm, and the classification accuracy for material and position reaches 100%. Our code is publicly available at https://drive.google.com/drive/folders/131M8zHzkvKUargplYRV 2oMuahUXq_-o1?usp=drive_link#