AC-YOLO: A lightweight ship detection model for SAR images based on YOLO11

Area of Science:

• Marine science and technology
• Remote sensing and geospatial analysis
• Artificial intelligence in environmental monitoring

Background:

• Synthetic Aperture Radar (SAR) is crucial for maritime applications due to its all-weather imaging.
• Existing SAR ship detection algorithms struggle with accuracy and computational demands.
• Challenges include varied target scales, indistinct features, and complex backgrounds.

Purpose of the Study:

• To develop a novel, lightweight SAR ship detection model (AC-YOLO) addressing accuracy and computational cost limitations.
• To enhance the detection of small targets and improve feature discrimination in complex SAR imagery.
• To provide an efficient solution for maritime surveillance on edge computing platforms.

Main Methods:

• Proposed AC-YOLO, a lightweight SAR ship detection model based on YOLOv11.
• Designed a lightweight cross-scale feature fusion module for adaptive multi-scale information integration.
• Developed a hybrid attention enhancement module combining convolutional and self-attention mechanisms.
• Introduced the Minimum Point Distance Intersection over the Union (MPDIoU) loss function for optimized bounding box regression.

Main Results:

• AC-YOLO reduced parameter count by 30.0% and computational load by 15.6% compared to the baseline YOLOv11.
• Achieved an average precision (AP) improvement of 1.2% on the SSDD dataset.
• Increased AP by 1.5% on the HRSID dataset.
• Demonstrated effective reconciliation of complexity and detection accuracy.

Conclusions:

• AC-YOLO offers a significant advancement in SAR ship detection, balancing efficiency and performance.
• The model's lightweight design makes it suitable for deployment on resource-constrained edge computing platforms.
• This research provides a practical solution for enhanced maritime security and management using SAR technology.