Three-Dimensional Complex Permittivity Reconstruction via Wave-Number Space Based Deep Neural Network for Microwave Breast Imaging

This study introduces a three-dimensional (3D) complex permittivity profile reconstruction using a deep neural network, where wave-number space data compression is applied to reduce the dimension of input data. Four-dimensional scattered data are converted into a 3D complex permittivity profile by integrating a 3D convolutional autoencoder and a multilayer perceptron. The reconstruction accuracy is further improved through efficient skin surface rejection preprocessing via a fractional derivative model. An experimental study, using simplified 3D breast phantom and an ultrawideband radar module shows that our proposed scheme provides accurate estimates for 3D reconstruction in terms of relative permittivity and conductivity.