A SAR Target Recognition Method via Combination of Multilevel Deep Features.

Area of Science:

• Remote Sensing
• Computer Vision
• Artificial Intelligence

Background:

• Synthetic Aperture Radar (SAR) imaging is crucial for target recognition.
• Accurate identification of targets in SAR imagery remains challenging due to factors like noise and occlusion.
• Deep learning and sparse representation offer complementary strengths for feature extraction and classification.

Purpose of the Study:

• To propose an effective method for SAR image target recognition by integrating multilevel deep features.
• To leverage the strengths of Residual Networks (ResNet) for feature learning and Joint Sparse Representation (JSR) for classification.
• To improve the overall performance and robustness of SAR target recognition systems.

Main Methods:

• Utilizing ResNet to extract multilevel deep features from SAR images.
• Clustering extracted features using similarity measures to form distinct feature sets.
• Classifying each feature set with JSR and combining results via weighted fusion.
• Validating the method on the MSTAR dataset under diverse conditions.

Main Results:

• The proposed method demonstrates superior performance in recognizing 10 types of target samples.
• Effective recognition was achieved under standard operating conditions, noise interference, and occlusion.
• The combination of ResNet and JSR significantly enhances feature extraction and classification accuracy.
• Experimental results confirm the method's effectiveness and robustness.

Conclusions:

• The integrated approach of ResNet and JSR offers a powerful solution for SAR image target recognition.
• The weighted fusion of JSR outputs from clustered deep features improves recognition accuracy.
• The method shows significant promise for real-world SAR applications requiring reliable target identification.