Advancing Underwater Image Enhancement Using Hybrid Deep Learning Models

Abstract

When attempting to make underwater images better, there are a lot of obstacles to overcome, including water's inherent color distortion, inadequate visibility, and light absorption and dispersion noise. The difficulties in accurately analyzing images caused by these problems apply to marine research, underwater robots, and environmental monitoring are only a few areas which are impacted. To address the limitations of existing picture enhancement techniques, this paper proposes a new hybrid deep learning CNN + GAN model for enhancing underwater photographs. The hybrid model extracts features with GANs and noise reduces with CNNs to produce better photos with better clarity and color correction. I train and validate the model using an underwater picture dataset and it defeats the state of the art approaches on visibility, color correctness, contrast and contrast. In quantitative and qualitative assessments our hybrid deep learning method yields superior generalization in different undersea environments relative to traditional methods. A suggested approach can potentially improve the underwater photography, marine setting data collecting with more precision, as well as better underwater exploration. This work now provides confidence in research progress into autonomous underwater exploration systems and underwater image processing.