FEAPN: Feature Enhancement and Alignment Pyramid Network for Underwater Object Detection

Underwater object detection plays a crucial role in the domain of marine engineering. Due to blur, uneven illumination and noise in underwater images, generic object detectors often fail to accurately detect underwater targets. Existing underwater object detection methods generally neglect the enhancement and refinement of multi-scale features, limiting further improvements in detection accuracy. In response to these challenges, we propose the Feature Enhancement and Alignment Pyramid Network (FEAPN), a novel underwater object detection framework. FEAPN consists of two key innovations. First, the Adaptive Feature Refinement Module (AFRM) is developed to adaptively enhance contextual features from complex backgrounds. Second, the Dual-path Feature Alignment Module (DFAM) is designed to align multi-scale features, utilizing cross-layer information to optimize feature representation. Extensive experiments demonstrate that FEAPN achieves state-of-the-art performance. Specifically, FEAPN achieves a 2.4% mAP improvement over the baseline and outperforms the current leading underwater detector by 1.2% mAP. Furthermore, the effectiveness of each component is validated through ablation studies.