High Efficiency Image Correction for Low Power Underwater Dron

In this paper, a high efficiency underwater image correction method based on the evaluation of structural similarity and real-time performance of a traditional FUnIEGAN algorithm is proposed. The proposed method can be implemented on an embedded GPU module (Jetson Xavier NX). The proposed method also enhanced the color correction performance by a data generation methodology. In addition, AMP is used to reduce computational redundancy, which enables fast operation. As a result, object detection results are improved, and FPS is increased to 41.12.