YOLOv8-DCW: Enhanced Object Detector Based on Deformable Convolution and Coordinate Attention

Underwater vision detection technology is essential for ocean exploration, driven by the growing need for accurate detection of underwater objects. In this study, we propose an enhanced underwater object detector that significantly improves detection accuracy. The model integrates DCNv3 with the C2f module, replacing the C2f module with C2f-Dv3 to better capture features of underwater objects in complex scenes and enhance the recognition of blurred and colorshifted objects. Integrating the Coordinate Attention into the original backbone network, in order to improve the detection of overlapping and occluded underwater objects. Additionally, the original network's CIoU is enhanced to WIoU v3 to improve the detection success rate of small underwater objects. Through experiments, our model YOLOv8-DCW achieved an mAP of 84.6% on the URPC2020 dataset, which is a 1.7% enhancement over the original YOLOv8n model, surpassing a range of YOLO models and the benchmark model YOLOv8n. These experimental results indicate that the improved model proposed in this paper demonstrates good applicability and generalization capability in detecting small underwater targets in complex environments.