2026-11-16T00:00:00-06:00
Loading Events

COMMITTEE CHAIR: Dr. Suxia Cui

COMMITTEE CO-CHAIR: Dr. Lujun Zhai

TITLE: INTEGRATION OF ATTENTION MECHANISMS IN YOLOV11 FOR IMPROVED UNDERWATER OBJECT DETECTION

ABSTRACT: Underwater ecosystems, covering over 70% of Earth’s surface, harbor diverse marine flora and fauna critical to environmental sustainability and biodiversity. Despite their significance, only approximately 5% of these aquatic environments have been explored, limiting our understanding of marine species distribution, behavior, and conservation needs. Automated object detection systems offer a promising solution for large-scale underwater monitoring, yet face significant challenges including degraded image quality, color distortion, low contrast, and complex backgrounds inherent to aquatic environments. This research investigates the integration of attention mechanisms with YOLOv11 for improved marine species detection using the DUO (Detecting Underwater Objects) dataset. The DUO dataset comprises 6,671 training images and 1,111 test images across four marine organism categories: Holothurian (sea cucumbers), Echinus (Sea urchins), Scallop, and Starfish, totaling 74,515 annotated instances. Initial baseline experiments with YOLOv11 achieved 81% Mean Average Precision (mAP) and 80% overall accuracy. To address the unique challenges of underwater object detection, particularly small object detection and feature degradation, we propose integrating Sequential Fusion Attention (SFA) mechanisms into the YOLOv11 architecture. The SFA module combines spatial and channel attention to enhance feature extraction while maintaining computational efficiency. Furthermore, the dataset includes several benthic organisms. The model is refined to highlight edge boundaries in these sea dwelling organisms. This proposed modification is termed the benthic Efficient Attention – Sequential Fusion Attention (BEA-SFA). Results show an improvement in overall mAP of over 80% and quicker inference time. This research aims to improve detection performance for marine species monitoring applications while maintaining real-time processing capabilities suitable for underwater robotic platforms.

Keywords: Object Detection, Attention Mechanisms, Sequential Fusion Attention

Room Location: Electrical and Computer Engineering Department Conference Room 315D

Share This Story, Choose Your Platform!

Go to Top