A Novel YOLOv5-Based Hybrid Underwater Target Detection Algorithm Combining with CBAM and CIoU

The algorithm for detecting underwater targets plays a critical role in facilitating the effective operation of an underwater robot in an aquatic environment. Currently, a wide array of land-based target detection methods are accessible. Nevertheless, the use of these algorithms in underwater target identification situations is challenging due to the intricate underwater environment, encompassing factors such as low light conditions, significant occlusion, wave interference, and other complexities. This study presents a unique approach for detecting underwater targets, utilizing the YOLOv5 framework. The integration of the CBAM attention mechanism module, featuring the PRelu activation function, with YOLOv5 is undertaken as an initial step to augment the network's capability in extracting intricate details. Additionally, the framework for target detection incorporates the CIo U Loss activation function and one cycle learning rate mode. These additions aim to expedite the convergence time of the network model and improve its detection precision. In conclusion, a novel approach for integrating characteristics is introduced, which effectively integrates more precise and detailed information. The experimental findings confirm the efficacy of the suggested framework in detecting submerged targets. The detection of small underwater targets is achieved with a mean Average Precision (mAP) of 83.9%, surpassing the performance of RCNN, Fast RCNN, and YOLOv5 target detection models.