Accurate and real-time visual detection algorithm for environmental perception of USVS under all-weather conditions

Owing to the intricate and ever-changing nature of the marine environment, traditional marine survey methods are subject to numerous limitations. Unmanned surface vehicles (USVs) have gained significant popularity for their role in automatically identifying and positioning targets in the ocean. To e...

Full description

Saved in:
Bibliographic Details
Published in:Journal of real-time image processing Vol. 21; no. 2; p. 36
Main Authors: Dong, Kaiyuan, Liu, Tao, Shi, Zhen, Zhang, Yang
Format: Journal Article
Language:English
Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.04.2024
Springer Nature B.V
Subjects:
Accuracy
Algorithms
Automatic vehicle identification systems
Classification
Computer Graphics
Computer Science
Datasets
Deep learning
Design
Frames per second
Image Processing and Computer Vision
Localization
Marine environment
Multimedia Information Systems
Neural networks
Object recognition
Optimization
Pattern Recognition
Perception
Real time
Signal,Image and Speech Processing
Surface vehicles
Target detection
Target recognition
Underwater exploration
Unmanned vehicles
Weather
Maritime dataset
Pyramid attention networks
Multi-scale feature fusion
Unmanned surface vehicles
Sea surface object detection
ISSN:1861-8200, 1861-8219
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Owing to the intricate and ever-changing nature of the marine environment, traditional marine survey methods are subject to numerous limitations. Unmanned surface vehicles (USVs) have gained significant popularity for their role in automatically identifying and positioning targets in the ocean. To enhance the environmental perception capabilities of USVs in complex marine environments, vision-based sea surface object detection algorithms have emerged as a crucial technological approach. In response to the unique challenges of sea surface object detection tasks and the various extreme situations encountered by USVs in real sea environments, YOLOv7 was chosen as the baseline model due to its excellent trade-off between speed and accuracy. We propose Efficient Multi-Scale Pyramid Attention Networks, which enable easy and rapid multi-scale feature fusion. Additionally, we improve the boundary box loss function. Building upon these optimizations, we develop a new detector called Marit-YOLO, which consistently achieves better accuracy and efficiency than the prior art across a wide spectrum of resource constraints. Marit-YOLO achieved a 5.0% increase in the average precision (AP) value on the independently collected Ocean Buoys Dataset. During generalization validation, the AP value also saw a 7.7% increase on the open-source Singapore Maritime Dataset. The Marit-YOLO algorithm achieves a real-time inference speed of 69 frames per second on a single RTX2080, enabling accurate and real-time target detection in complex sea environments.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:1861-8200
1861-8219
DOI:10.1007/s11554-024-01417-9