Underwater Object Detection Based on Image Enhancement and Multi-Branch Structure

Underwater object detection plays an important role for underwater intelligent applications. However, optical-based underwater images often suffer from color bias, haze-like, small scale object and occlusions, resulting in lower detection accuracy for detectors. To address the above problems, an underwater object detection network based on image enhancement and multibranch structure is proposed. Concretely, a parameter-free color correction attention module is first presented to adjust the color cast of underwater images, providing detection-favorable features to the network. In addition, a backbone network and a neck network are constructed based on the attention mechanism and multi-branch structure to improve the detection ability of the network for small and occluded objects. The decoupled head is further introduced to implement the regression and classification tasks. Extensively qualitative and quantitative experiments show that the proposed method achieves a maximum AP of $53.2 \%$ at 280 FPS in the UODD dataset, outperforming other mainstream methods.