A Multi-Scale Convolutional Hybrid Attention Residual Network for Enhancing Underwater Image and Identifying Underwater Multi-Scene Sea Cucumber

Abstract

At present, the use of underwater robots to replace underwater manual work is a future development direction. The complex and changeable underwater environment brings great difficulties to the operation of robots. In order to improve the problem of color distortion and degradation of sea cucumber images and to increase the accuracy and stability of sea cucumber detection, this letter proposes a data-driven multi-scale convolutional hybrid attention residual enhancement network (MCRNet). Firstly, a gain factor that considers the differences of each color channel is used to neutralize the color bias of underwater images. Secondly, a multi-scale convolutional fusion module is proposed to extract shallow feature information at different scales. Multiple residual blocks and hybrid attention module are constructed to focus on extracting deep features in the image. The overall residual connectivity of the network helps to retain the shallow information. The experimental results show that compared with the unprocessed four types of sea cucumber images, the method proposed in this letter improves 63.96% and 44.50% in the image quality evaluation metrics UCIQE and UIQM, respectively, and the average detection accuracy of sea cucumbers is improved by 4.7%.