关键词: 关键词: YOLOv8, 水下目标检测, 特征融合, FCA注意力, WIoU v3损失函数

Abstract: Abstract: Due to the unique characteristics of underwater creatures, underwater images usually exit many small targets being hard to detect which often overlap with each other. In addition, the presence of light absorption and scattering in underwater environment can cause underwater images to occur color shift and blurriness. Aim to overcome those challenges, this paper propose an enhanced underwater target detection algorithm, namely WCA-YOLOv8. Firstly, WCA-YOLOv8 adds the Feature Fusion Module (FFM) to improve the focus on spatial dimension to improve the recognition of blurry and color-shifted targets. At second, WCA-YOLOv8 adds the FCA module to enhance the feature extraction for overlapped and occluded underwater targets. At third, WCA-YOLOv8 uses CIoU loss function to replace WIoU v3 loss function to strengthen the ability to detect small size targets. At last, WCA-YOLOv8 adds newly designed DEM module to preserve context information during feature extraction more effectively. Experimental results show that WCA-YOLOv8 can achieves 75.8% and 88.6% detection accuracy and 60FPS and 57FPS detection speeds on RUOD and URPC data sets, respectively. Compared with other evaluated state-of-the-art object detection algorithms, WCA-YOLOv8 can achieve higher detection accuracy with faster detection speed.

Key words: Keywords: YOLOv8, Underwater target detection, Feature Fusion, FCA attention, WIoU v3 loss function