Underwater hyperspectral image recovery based on a single chromatic aberration blur image using deep learning

Abstract

Hyperspectral imaging technology can capture the spatial information and spectral information in the scene, so it has a wide range of application prospects in the fields of remote sensing and target recognition. The underwater environment will absorb or scatter the light beam emitted by the light source, which makes it difficult for the light sensing element to perceive all the spectral information of the target, resulting in problems such as low resolution, high complexity, and long exposure time of underwater hyperspectral imaging. We propose a novel underwater hyperspectral imaging method, using a self-developed lens with longitudinal chromatic aberration in front of a monochrome camera. This device captures a single frame of chromatic aberration blur image at a fixed focus position (550nm) to realize the recovery of 146 bands of hyperspectral image in the range of 430nm-720nm. In this paper, the U-NET network in the convolutional neural network is implemented to complete the training process from a single chromatic aberration blurred image to a hyperspectral image through the deep learning method, and achieve good experimental results in the laboratory. The results show that this method is feasible and can effectively extract hyperspectral images from monochromatic chromatic aberration blurred images.