Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Rethinking SAR ATR: A Target-Aware Frequency-Spatial Enhancement Framework with Noise-Resilient Knowledge Guidance

Yansong Lin, Zihan Cheng, Jielei Wang, Guoming Lua, Zongyong Cui

Abstract

Synthetic aperture radar automatic target recognition (SAR ATR) is of considerable importance in marine navigation and disaster monitoring. However, the coherent speckle noise inherent in SAR imagery often obscures salient target features, leading to degraded recognition accuracy and limited model generalization. To address this issue, this paper proposes a target-aware frequency-spatial enhancement framework with noise-resilient knowledge guidance (FSCE) for SAR target recognition. The proposed framework incorporates a frequency-spatial shallow feature adaptive enhancement (DSAF) module, which processes shallow features through spatial multi-scale convolution and frequency-domain wavelet convolution. In addition, a teacher-student learning paradigm combined with an online knowledge distillation method (KD) is employed to guide the student network to focus more effectively on target regions, thereby enhancing its robustness to high-noise backgrounds. Through the collaborative optimization of attention transfer and noise-resilient representation learning, the proposed approach significantly improves the stability of target recognition under noisy conditions. Based on the FSCE framework, two network architectures with different performance emphases are developed: lightweight DSAFNet-M and high-precision DSAFNet-L. Extensive experiments are conducted on the MSTAR, FUSARShip and OpenSARShip datasets. The results show that DSAFNet-L achieves competitive or superior performance compared with various methods on three datasets; DSAFNet-M significantly reduces the model complexity while maintaining comparable accuracy. These results indicate that the proposed FSCE framework exhibits strong cross-model generalization.

Rethinking SAR ATR: A Target-Aware Frequency-Spatial Enhancement Framework with Noise-Resilient Knowledge Guidance

Abstract

Synthetic aperture radar automatic target recognition (SAR ATR) is of considerable importance in marine navigation and disaster monitoring. However, the coherent speckle noise inherent in SAR imagery often obscures salient target features, leading to degraded recognition accuracy and limited model generalization. To address this issue, this paper proposes a target-aware frequency-spatial enhancement framework with noise-resilient knowledge guidance (FSCE) for SAR target recognition. The proposed framework incorporates a frequency-spatial shallow feature adaptive enhancement (DSAF) module, which processes shallow features through spatial multi-scale convolution and frequency-domain wavelet convolution. In addition, a teacher-student learning paradigm combined with an online knowledge distillation method (KD) is employed to guide the student network to focus more effectively on target regions, thereby enhancing its robustness to high-noise backgrounds. Through the collaborative optimization of attention transfer and noise-resilient representation learning, the proposed approach significantly improves the stability of target recognition under noisy conditions. Based on the FSCE framework, two network architectures with different performance emphases are developed: lightweight DSAFNet-M and high-precision DSAFNet-L. Extensive experiments are conducted on the MSTAR, FUSARShip and OpenSARShip datasets. The results show that DSAFNet-L achieves competitive or superior performance compared with various methods on three datasets; DSAFNet-M significantly reduces the model complexity while maintaining comparable accuracy. These results indicate that the proposed FSCE framework exhibits strong cross-model generalization.
Paper Structure (19 sections, 12 equations, 6 figures, 11 tables)

This paper contains 19 sections, 12 equations, 6 figures, 11 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. SAR ATR
  4. Knowledge Distillation
  5. The Proposed Method
  6. Data Preprocessing
  7. Frequency-Spatial Collaborative Enhancement Framework
  8. Student Model and Shallow Feature Enhancement
  9. Teacher Model and KD Design
  10. Experiments
  11. Dataset Details
  12. Experimental Setup
  13. Comparison with Representative Methods
  14. FSCE Framework Transferability Experiment
  15. Hierarchical Sensitivity Experiment
  16. ...and 4 more sections

Figures (6)

  • Figure 1: The overall architecture of our proposed method. DSAFNet-L and DSAFNet-M use ResNet18 and ShuffleNetV2 0.25x as the backbone network respectively, and use FSCE for focus area guidance to achieve spatial-frequency domain adaptive joint feature selection enhancement. The implementation path of each part is given.
  • Figure 2: ResNet18 is used to display feature maps of different depths and their feature maps after passing through the DSAF module. The gray image on the left is the image in MSTAR, and heatmaps on the right. heatmaps from left to right are Pre layer, Layer1, layer2, Layer3, Layer4.
  • Figure 3: Grad-CAM is used to compare the heat maps of the interest areas of different layers of each model. From top to bottom, the models are ResNet101 pre-trained model, ResNet18 pre-trained model, ResNet18 model without pretraining but with the same training configuration as DSAFNet-L, and DSAFNet-L.
  • Figure 4: MSTAR 10-classification dataset.
  • Figure 5: OpenSARShip and FUSARShip 4-classification datasets
  • ...and 1 more figures