Research of Underwater Image Enhancement Techniques Utilizing Deep Convolutional Networks

In response to the issues of blurriness, color bias, and fogging encountered when modern underwater robots or industrial equipment capture underwater images, this paper proposes an innovative underwater image enhancement algorithm. These issues significantly reduce work efficiency. The algorithm employs a novel three-branch deep convolutional structure, designed to enhance the visual quality and the efficiency of information extraction from degraded images by utilizing diverse deep convolutional modules. Firstly, the core component of the algorithm introduces a deep feature extraction module, specifically designed to capture both deep and shallow features of degraded images, thereby enhancing the efficiency of subsequent processing and applications. Secondly, the auxiliary branch integrates a spatial depth extraction module to delve into the spatial structure and color details of the image, restoring its natural visual perception. Finally, to address the common issue of fogging in underwater images, the third branch introduces an attention mechanism to enhance the network's flexible handling of image details. Experimental results show that our proposed method outperforms existing classic and emerging algorithms in multiple aspects. In terms of objective evaluation metrics, our method achieves top performance, demonstrating its excellent performance and applicability in real underwater image enhancement tasks.