AWF-YOLO: enhanced underwater object detection with adaptive weighted feature pyramid network

Abstract

Underwater scenarios are influenced by various factors such as light attenuation, scattering, and absorption, which degrade the quality of images and pose significant challenges for underwater object detection in marine research and ocean engineering. To address these challenges, we propose a novel adaptive-weight feature detection framework based on YOLOv8, called AWF-YOLO, designed to detect objects in turbid underwater scenarios accurately. AWF-YOLO incorporates several key components to improve detection performance. Firstly, a novel adaptive-weight feature pyramid network is introduced to facilitate the fusion of multi-scale feature semantics. In addition, an adaptive-weight feature extraction module is proposed to enhance underwater object detection by capturing relevant and discriminative information to enhance feature extraction further. We integrate a dedicated small object detection head into the detection network to overcome the challenges associated with detecting small objects in complex underwater scenarios. This component focuses on effectively identifying and localizing small objects, leading to improved overall detection accuracy. Extensive experiments conducted on the detection underwater objects dataset demonstrate that the proposed AWF-YOLO achieves significant performance improvements, thus making it highly suitable for complex and dynamic underwater scenarios.