Super resolution for microwave imaging: A deep learning approach

We propose a method to significantly improve the spatial resolution for microwave imaging. Conventional inverse methods for microwave imaging produce images at varying levels of resolution but none of them are at the resolution sufficiently fine to be useful for a complex real-world problem, mainly because of limited availability of independent measurements. We ease the problem by providing additional information through learning. We incorporate learning using a convolutional neural network in the second stage of our proposed two-staged approach, where the first stage is a non-linear inversion approach. Our method can be used with any conventional method and can boost the resolution in all dimensions. The applicability of our method is demonstrated for a 2D microwave imaging problem for an upscale factor of 3. The results show that the proposed method can produce a better detailed higher resolution image.