Dynamic Correction and Deep Convolution-based Underwater Image Enhancement Algorithm for Underwater Robots

Addressing the issue of limited work efficiency for intelligent robots in extreme and complex underwater environments such as the seabed, we propose an innovative underwater image enhancement method that combines dynamic color correction and deep convolutional technology. Firstly, the classical Gray World color correction algorithm has been improved to more robustly handle the intricate color cast challenges in underwater environments. Following this, Dense Residual Blocks are introduced to deepen the architecture of the convolutional neural network. By incorporating fully connected layers, the transmission efficiency and accuracy of information within the network are enhanced. Additionally, we have designed a Parallel Attention Module that reduces the network's parameter burden while granting it flexible processing capabilities. Furthermore, a Global Compensation Module is proposed, which further boosts the network's adaptability to complex variations in underwater environments. Experimental results demonstrate that our proposed method outperforms relevant comparison algorithms in both subjective and objective tests, more effectively resolving underwater degraded images and enhancing the work efficiency of underwater robots.