Vehicle Localization Method in Complex SAR Images Based on Feature Reconstruction and Aggregation.

Area of Science:

• Remote Sensing
• Computer Vision
• Artificial Intelligence

Background:

• High-resolution synthetic aperture radar (SAR) images present challenges for vehicle detection due to small target size and complex backgrounds.
• Existing deep learning methods struggle with extracting high-quality vehicle features, impacting localization accuracy.

Purpose of the Study:

• To propose an advanced vehicle localization method for SAR images.
• To enhance feature extraction and representation quality for improved detection.

Main Methods:

• Utilized a backbone network with a space-channel reconfiguration module (SCRM) for SAR-specific feature extraction.
• Implemented a progressive cross-fusion mechanism (PCFM) to integrate multi-view features from different layers.
• Employed a rotating box detection head for precise vehicle localization.

Main Results:

• The proposed method significantly improves vehicle localization accuracy in complex SAR image datasets.
• Achieved state-of-the-art performance, demonstrating superior effectiveness.

Conclusions:

• The feature reconstruction and aggregation method with rotating boxes offers a robust solution for vehicle localization in SAR imagery.
• The integration of SCRM and PCFM enhances feature representation, leading to reduced false alarms and missed detections.