MSADNet: Multistage SAR Aircraft Target Detection Network

Aircraft target detection in synthetic aperture radar (SAR) images is challenging due to discrete scattering points and complex environments. With neural network advancements, this task has become more feasible, although the existing networks often have too many parameters for performance-constraine...

Full description

Saved in:
Bibliographic Details
Published in:IEEE geoscience and remote sensing letters 2025, Vol.22, p.1-5
Main Authors: Wang, Xi, Xu, Wei, Huang, Pingping, Tan, Weixian
Format: Article
Language:English
Subjects:
Aggregation
Aircraft
Aircraft detection
Aircraft performance
Aircraft target detection
Asymmetry
Computational efficiency
Computational modeling
Convolution
Drone aircraft
Feature extraction
lightweight neural networks
Modules
Neck
Neural networks
Parameters
Radar detection
Radar imaging
Radar polarimetry
SAR (radar)
Scattering
Synthetic aperture radar
synthetic aperture radar (SAR)
Target detection
Training
Unmanned aerial vehicles
Citations: Items that this one cites
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Aircraft target detection in synthetic aperture radar (SAR) images is challenging due to discrete scattering points and complex environments. With neural network advancements, this task has become more feasible, although the existing networks often have too many parameters for performance-constrained devices like drones or sacrifice performance to reduce parameter count. To address this, we propose a lightweight multistage SAR aircraft target detection network, termed MSADNet, featuring a novel multistage asymmetric aggregation network (MAAN) module. The MAAN module enhances feature extraction through an improved asymmetrical bottleneck network and introduces a multiorder attention mechanism to focus on critical features. By carefully managing the shortest and longest gradient paths, the MAAN module achieves an organic integration of both, significantly improving network performance. The backbone network is constructed using the MAAN module, and the neck network employs a feature pyramid network (FPN) + PAN structure, integrating the MAAN module with a coordinate attention (CA) module. An anchor-free detection head is used for aircraft target detection. Testing on the SAR-AIRcraft-1.0 dataset demonstrates that MSADNet achieves a mean average precision (mAP) of 96.15% with only 4.88 M parameters. Compared to YOLOv8-s, MSADNet improves mAP by 0.61 percentage points while reducing the parameter count by 6.29 M, indicating high-performance detection with a low parameter count.
ISSN:1545-598X
1558-0571
DOI:10.1109/LGRS.2024.3521650