LW-YOLO11: A Lightweight Arbitrary-Oriented Ship Detection Method Based on Improved YOLO11.

Area of Science:

• Computer Vision
• Remote Sensing
• Artificial Intelligence

Background:

• Arbitrary-oriented ship detection in remote sensing images is challenging due to high resolution, poor clarity, and scale variations.
• Existing methods often struggle to balance high accuracy and detection speed.

Purpose of the Study:

• To design a lightweight and efficient module for high-precision, arbitrary-oriented ship detection in remote sensing images.
• To improve the performance of the YOLOv11 network for this specific task.

Main Methods:

• Developed a lightweight and efficient multi-scale feature dilated neck module integrated into the YOLOv11 network.
• Utilized multi-scale dilated attention to capture semantic details.
• Employed cross-stage partial networks for spatial-semantic information interaction.
• Introduced the GSConv module to preserve semantic information during feature transmission.

Main Results:

• The proposed method demonstrates a lightweight structure and high accuracy.
• Achieved superior ship detection performance compared to state-of-the-art methods.
• On the HRSC2016 dataset, improvements of 3.1% mAP@0.5 and 3.3% mAP@0.5:0.95 were observed compared to YOLOv11n.
• On the MMShip dataset, improvements of 1.9% mAP@0.5 and 1.3% mAP@0.5:0.95 were achieved compared to YOLOv11n.

Conclusions:

• The designed module effectively enhances arbitrary-oriented ship detection in remote sensing imagery.
• The method offers a promising solution for accurate and efficient ship detection, outperforming existing approaches.