Super-Resolution 3-D Microwave Imaging of Objects With High Contrasts by a Semijoin Extreme Learning Machine

This article proposes a semijoin extreme learning machine (SJ-ELM) for super-resolution 3-D microwave imaging of objects with high contrasts. The proposed scheme develops a shallow neural network structure with the semijoin strategy to convert the scattered field data into two output channels, namely, the permittivity and the conductivity of objects, respectively. The semijoin strategy can decrease the inner matrix dimensions to reduce the computational burden for 3-D super-resolution imaging, so it is employed to connect between the nodes of the hidden layer and the output layer. The imaging performance of the proposed SJ-ELM and the conventional variational Born iterative method (VBIM) is first compared for imaging objects with different electrical sizes and contrasts, and then, different targets of imaging resolution are designed to evaluate both solvers. The proposed SJ-ELM is also assessed for imaging objects with high contrasts and experimental data and is demonstrated to have superior super-resolution imaging capabilities for high-contrast 3-D objects.