DPT-YOLO: An Improved Underwater Object Detection Model Based on YOLO

Underwater object detection is vital for marine exploration, environmental monitoring, and military applications. However, the complexity of underwater environments poses significant challenges for detection technologies. This study introduces DPT-YOLO, a novel network architecture that enhances the...

Full description

Saved in:
Bibliographic Details
Main Authors: Huang, Kexiang, Xu, Minghe, Guo, Te, Chen, Fusheng, Wei, Chung-Lun, Chu, Chiawei
Format: Conference Proceeding
Language:English
Subjects:
Computational modeling
deep learning
Feature extraction
Head
Network architecture
object detection
Real-time systems
Robustness
Target recognition
Transformers
underwater images analysis
Wireless communication
YOLO
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Underwater object detection is vital for marine exploration, environmental monitoring, and military applications. However, the complexity of underwater environments poses significant challenges for detection technologies. This study introduces DPT-YOLO, a novel network architecture that enhances the traditional YOLO framework with a dual-path transformer module, improving geometric feature extraction for target detection. It also leverages multi-scale feature fusion to better detect small targets by capturing both local and global features, enhancing robustness. The YOLO detection head ensures efficient real-time performance. Experimental results demonstrate that DPT-YOLO significantly outperforms traditional models in accuracy, offering valuable insights for underwater object detection and practical applications.
ISSN:2377-8512
DOI:10.1109/CSRSWTC64338.2024.10811580