Scholarlink

Recently, there has been significant progress in Electrical Capacitance Tomography (ECT) image reconstruction, yet there remains room for enhancing both speed and quality in these reconstructions. Deep Learning (DL) offers a promising avenue due to its capacity to handle intricate nonlinear functions. This study introduces DL-based ECT image reconstruction model based on Residual UNet and termed as ECT_ResUNet. A sizable dataset containing permittivity distribution and corresponding capacitance measurements was created to train and evaluate the proposed model. The ECT_ResUNet processes modulated capacitance vectors, representing them as \(66\times 66\) images. The model’s effectiveness and practicality were assessed using unseen samples, demonstrating precise image reconstruction with an average Image Error (IE) of 0.0918 and Correlation Coefficient (CC) of 0.9527.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Advancing ECT Imaging: Residual UNet Deep Neural Network Approach

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Wael Deabes,
  • Fahmi Khalifa,
  • Alaa E. Abdel-Hakim,
  • Kheir Eddine Bouazza

摘要

Recently, there has been significant progress in Electrical Capacitance Tomography (ECT) image reconstruction, yet there remains room for enhancing both speed and quality in these reconstructions. Deep Learning (DL) offers a promising avenue due to its capacity to handle intricate nonlinear functions. This study introduces DL-based ECT image reconstruction model based on Residual UNet and termed as ECT_ResUNet. A sizable dataset containing permittivity distribution and corresponding capacitance measurements was created to train and evaluate the proposed model. The ECT_ResUNet processes modulated capacitance vectors, representing them as \(66\times 66\) images. The model’s effectiveness and practicality were assessed using unseen samples, demonstrating precise image reconstruction with an average Image Error (IE) of 0.0918 and Correlation Coefficient (CC) of 0.9527.