
ASFF-Det: Adaptive Space-Frequency Fusion Detector for Object Detection in SAR Images
Author(s) -
Zuohui Chen,
Hao Wu,
Wei Wu,
Haiping Yang,
Liao Yang,
Kun Li
Publication year - 2025
Publication title -
ieee journal of selected topics in applied earth observations and remote sensing
Language(s) - English
Resource type - Magazines
SCImago Journal Rank - 1.246
H-Index - 88
eISSN - 2151-1535
pISSN - 1939-1404
DOI - 10.1109/jstars.2025.3593313
Subject(s) - geoscience , signal processing and analysis , power, energy and industry applications
Synthetic Aperture Radar (SAR) is essential for high-resolution Earth observation, offering all-weather capabilities. However, SAR target detection is challenging due to its coherent imaging mechanism, which introduces spatial-frequency correlations in target scattering properties. Unlike optical images, SAR images lack well-defined edges and continuous textures, making spatial feature extraction alone insufficient for robust recognition. The frequency domain provides complementary structural information, where low-frequency components capture overall shape and backscatter characteristics, while high-frequency components highlight fine details but are dominated by speckle noise. Balancing spatial and frequency features is thus crucial for accurate detection and noise suppression. To address this, we propose Adaptive Space-Frequency Fusion Detector (ASFF-Det), which integrates spatial and frequency features for SAR target detection. Our Adaptive Space-Frequency Fusion Convolution (ASFF-Conv) module integrates frequency domain transformations with spatial feature extraction, enabling adaptive multi-domain feature fusion. To mitigate high-frequency noise, our Frequency-Intensity Sorted Attention (FSA) mechanism prioritizes low-frequency structural information. Additionally, our High-Low Frequency Fusion (HLFF) module enhances multi-scale target recognition by frequency decomposition and feature enhancement. Extensive experiments on the SAR-AIRcraft-1.0, HRSID, and SSDD datasets demonstrate ASFF-Det's superiority, achieving 68.5%, 91.8%, and 96.4% mAP $_{50}$ , respectively, outperforming state-of-the-art methods. Our code is released at https://github.com/bleudeprusse/ASFF-Det
Accelerating Research
Robert Robinson Avenue,
Oxford Science Park, Oxford
OX4 4GP, United Kingdom
Address
John Eccles HouseRobert Robinson Avenue,
Oxford Science Park, Oxford
OX4 4GP, United Kingdom