Enhancing YOLOv8n with Multiple Attention and MRV Module for Efficient Deep-Sea Pipeline Target Detection

Deep-sea pipeline target detection is a critical component in enhancing the efficiency of ocean resource exploitation. Traditional methods for underwater pipeline detection encounter numerous challenges, such as low brightness, high blur, and interference, which are further amplified in the deep-sea environment, and these shortcomings are further magnified in deep-sea environments. To address these issues, this paper introduces an enhanced YOLOv8n model (MRVS-YOLOv8n) aimed at improving detection performance. First, Squeeze-and-Excitation (SE) Attention is employed to augment the capabilities of the backbone network. Subsequently, we propose a novel neck network module, termed the MRV module, which replaces the C2f module and fully leverages contextual information to enhance global detection performance. Additionally, Shuffle Attention is incorporated, and the detection head is expanded to increase the model’s diversity in detections. Finally, we adopt the WIoUv3 loss function as a replacement for the CIoU loss function to bolster the model’s robustness against spatial transformations and scale variations. The results from model comparisons and ablation experiments validate the superiority of our improved model on test sets; specifically. The proposed MRVS-YOLOv8n framework demonstrates a remarkable mean Average Precision (mAP) of 98.7% and an improvement of 6.4% over the base model, underscoring its efficacy in addressing complex underwater scenarios. This advancement not only reduces resource wastage but also enhances deep-sea pipeline management practices—providing a novel solution within this field.