Short-wave infrared polarimetric image reconstruction using a deep convolutional neural network based on a high-frequency correlation.

Abstract

Imaging in visible and short-wave infrared (SWIR) wavebands is essential in most remote sensing applications. However, compared to visible imaging cameras, SWIR cameras typically have lower spatial resolution, which limits the detailed information shown in SWIR images. We propose a method to reconstruct high-resolution polarization SWIR images with the help of color images using the deep learning method. The training dataset is constructed from color images, and the trained model is well suited for SWIR image reconstruction. The experimental results show the effectiveness of the proposed method in enhancing the quality of the polarized SWIR images with much better spatial resolution. Some buried spatial and polarized information may be recovered in the reconstructed SWIR images.