CVNN-Based Microwave Imaging Approach

Signal processing involved in microwave imaging typically requires complex value representations. Nevertheless, most deep learning architectures rely on real value operations, thus limiting their capacity to learn complex value functions effectively. The present work focuses on the integration of deep learning and conventional Born Iterative method, with a specific emphasis on the use of a Complex Valued Neural Network for the reconstruction of dielectric profiles and the identification of breast tumors. Validation results demonstrate that the proposed approach works similarly to a standard Real Valued Neural Network. However, the main advantage is related to the direct operation in the complex value domain, enabling it to effectively capture both the magnitude and phase information. Consequently, the proposed approach is able to reconstruct both the real permittivity as well as the conductivity values in real time.