Abstract
The field of underwater object detection has gained significant attention recently, yet it remains challenged by issues like turbid water conditions and limited lighting, despite the rapid development in deep learning algorithms. In this paper, we try to provide a novel hybrid underwater object detection framework, named Efficient-YOLOv8, designed to address confusion-scale, low-resolution and small targets. In particular, our focus lies on the detection of cuttlefish, an underwater mollusc species known for its irregular postures. In our self-collected dataset, our proposed approach demonstrates remarkable results, achieving a precision of 94.18%, recall of 97.40%, and mAP of 98.18%.