YOLOv8-BFDS: A Framework for Underwater Object Detection Based on Feature Enhancement and Fusion

Object detection plays a critical role in various practical applications, but it is often hindered by challenges such as insufficient lighting, blurred imaging, and complex background interference. This paper proposes an enhanced YOLOv8 model, YOLOv8-BFDS, which integrates three key optimizations: the DCNv2 module dynamically adjusts the receptive field to better capture deformed and occluded objects; the E-SEModule combines channel and spatial attention mechanisms to enhance the detection of small and low-contrast objects; and the Concat_BiFPN module optimizes multi-scale feature fusion through bidirectional connections and adaptive weighting. Experimental results demonstrate significant improvements over the original YOLOv8, with precision increasing by 2.1% (from 0.95006 to 0.97011), recall by 19.1% (from 0.7985 to 0.98964), mAP50 by 14.2% (from 0.86825 to 0.99149), and mAP50-95 by 34.5% (from 0.62178 to 0.83604). These advancements demonstrate the model's superior robustness and accuracy in object detection tasks.