An Efficient Lightweight SAR Ship Target Detection Network with Improved Regression Loss Function and Enhanced...

Area of Science:

• Computer Science
• Artificial Intelligence
• Remote Sensing

Background:

• Ship identification in Synthetic Aperture Radar (SAR) images is challenging due to dense targets, complex backgrounds, and small object sizes.
• Current deep learning target detection algorithms often feature complex, large neural networks, hindering practical application speed and efficiency.

Purpose of the Study:

• To propose an efficient and lightweight deep learning network, Efficient-YOLO, for improved ship detection in SAR imagery.
• To enhance the localization accuracy, model convergence, feature extraction, and generalization performance for ship detection.

Main Methods:

• Introduction of a novel regression loss function, ECIOU, to improve bounding box accuracy and convergence.
• Development of the SCUPA module to enhance feature information multiplexing and model generalization.
• Proposal of the GCHE module to strengthen the network's feature extraction capabilities.

Main Results:

• Efficient-YOLO demonstrates superior performance compared to state-of-the-art algorithms on the SSDD and HRSID ship datasets.
• The proposed method achieves high accuracy and recall while maintaining a faster detection speed.
• Efficient-YOLO exhibits reduced model complexity and smaller model size, facilitating practical deployment.

Conclusions:

• Efficient-YOLO offers an effective and efficient solution for ship detection in complex SAR images.
• The novel components (ECIOU, SCUPA, GCHE) contribute significantly to the improved performance.
• The lightweight design makes Efficient-YOLO suitable for real-world applications requiring rapid and accurate ship identification.