Generative model-based image reconstruction from back-scattered data in computational microwave imaging

A novel cGAN is leveraged to achieve image restoration, where both the condition and the input are the CI-based microwave back-scattered measurements. The proposed cGAN consists of two parts. One is defined as the generator, which is leveraged to achieve image restoration. The other one is known as the discriminator, which serves the purpose of optimizing the generator by judging the similarity of the estimation and the ground truth. The training samples and testing samples are randomly selected from the open-source dataset MNIST. During the optimization process, the generator is adversarial with the discriminator. The optimized generator can retrieve high-fidelity image reconstructions directly from the CI backscattered measurements, eliminating the need for the computationally expensive image reconstruction step required by conventional imaging techniques. This contributes to the successful reconstruction of the scene images by deep learning methods. The performance of the proposed approach and its efficacy are confirmed by numerical simulations.