Underwater Image Enhancement and Super Resolution based on Deep CNN Method

The significance of exploration of underwater resources in the development and use of underwater autonomous operations are becoming more essential to protect against the risk of high-pressure deep-sea environments. To operate underwater autonomously intelligence computer vision will be the key technology. In a submerged environment light is weak and poor-quality image enhancement, used as pre-processing procedures, are required to enable underwater vision. The paper presents a mix that enhances images using CNN and CBAM, super resolution. This paper proposes to utilize attenuation module for adaptive residual learning to filter out the features which are irrelevant from the previous layers. It should be mentioned that both wavelength-driven multi-contextual design and attentive residual learning are not proposed for UIR. This paper presented a detailed study on how the performances of a few high-level vision tasks, such as diver's 2D pose estimation and underwater semantic segmentation, have been improved when presented with the enhanced images produced by Deep Wave Net.