Trial of Deep Learning for Image Reconstruction of Lens-Less Microwave Holography

We perform the principal verification of reconstructing object surface images by using deep learning. Using the deep learning neural network based on convolutional neural networks, simple object surface images with 128 x 128 pixels are reasonably reconstructed with up-converting from rough microwave signal images with 16 x 16 pixels. The model captures large structural features of the object surface images even with small number of training data. As the number of training data increases, it captures small structures of objects. It is also found that noises of input signal images affect reconstructions of small structures of objects. (C) 2022 The Japan Society of Plasma Science and Nuclear Fusion Research