U-Net-Based Multispectral Image Generation From an RGB Image

Multispectral images have lower spatial resolution than RGB images. It is difficult to obtain multispectral with both high and spectral because of expensive capture setup sophisticated acquisition processes. In this paper, we propose a deep neural network structure based on U-Net convert ordinary into resolution. Our variant not only preserves detailed features images, but also promotes the fusion different feature scales, enhancing quality image generation. Apart from training stage, our proposed method does require low-resolution as do some earlier learning-based methods; can be obtained using corresponding high-resolution We employ Inception block achieve richer loss function optimize non-local features. algorithm achieves state-of-the-art visual effects quantitative measurements such RMSE rRMSE several public datasets.