Sar Ship Detection Based on Convnext with Multi-Pooling Channel Attention and Feature Intensification Pyramid Network

Area of Science:

• Remote Sensing
• Computer Vision
• Artificial Intelligence

Background:

• Convolutional Neural Networks (CNNs) have advanced Synthetic Aperture Radar (SAR) ship detection.
• Existing algorithms face limitations in multiscale feature generation, false alarm suppression, and enhancing shallow feature semantics.
• Weakened semantic information in top-level feature maps due to channel reduction impacts detection performance.

Purpose of the Study:

• To address limitations in current SAR ship detection algorithms.
• To improve the accuracy and robustness of ship detection in SAR images.
• To enhance the detection of small ships and reduce false alarms.

Main Methods:

• Utilized Convnext as a backbone for high-quality multiscale feature map generation.
• Introduced Multi-Pooling Channel Attention (MPCA) to suppress false alarms and optimize feature maps.
• Developed Feature Intensification Pyramid Network (FIPN) and Top-Level Feature Intensification (TLFI) to enhance semantic information in feature maps.

Main Results:

• The proposed approach demonstrated superior performance on the SAR Ship Detection Dataset (SSDD).
• Achieved an overall Average Precision (AP) of 95.6% on the SSDD.
• Improved accuracy by at least 1.7% compared to existing advanced methods.

Conclusions:

• The novel approach effectively overcomes limitations in existing SAR ship detection techniques.
• The combination of Convnext, MPCA, FIPN, and TLFI significantly enhances detection accuracy and reliability.
• The method shows strong potential for practical applications in maritime surveillance.