LPDNet: A Lightweight Network for SAR Ship Detection Based on Multi-Level Laplacian Denoising

Area of Science:

• Remote Sensing
• Artificial Intelligence
• Computer Vision

Background:

• Maritime situational awareness relies heavily on accurate ship detection using Synthetic Aperture Radar (SAR).
• Deep learning models offer significant advantages for SAR ship detection but often struggle to balance computational efficiency with high accuracy.
• Existing methods lack a unified approach to address noise and feature representation challenges in SAR imagery.

Purpose of the Study:

• To propose an end-to-end, lightweight algorithm for SAR ship detection.
• To enhance the accuracy and efficiency of SAR ship detection models.
• To develop a method that effectively balances model size and detection performance.

Main Methods:

• Introduced a multi-level Laplacian pyramid denoising network (LPDNet) for adaptive, supervised denoising using Convolutional Neural Networks (CNNs).
• Implemented channel modeling to integrate spatial and frequency domain information for enhanced feature representation.
• Integrated the Convolutional Block Attention Module (CBAM) into the Yolox-tiny framework for improved feature fusion and highlighting.

Main Results:

• The proposed LPDNet achieved 97.14% Average Precision (AP) at 24.68 FPS on the SSDD dataset.
• On the AIR SARShip-1.0 dataset, LPDNet attained 92.19% AP at 23.42 FPS.
• The model demonstrated exceptional lightweightness with only 5.1 million parameters.

Conclusions:

• LPDNet offers a highly accurate and efficient solution for SAR ship detection.
• The algorithm effectively addresses the trade-off between model size and performance.
• The proposed method represents a significant advancement in lightweight deep learning for maritime surveillance.