A Novel Method for Recognizing Space Radiation Sources Based on Multi-Scale Residual Prototype Learning Network.

Area of Science:

• Space situational awareness
• Electromagnetic signal analysis
• Deep learning applications

Background:

• Space target recognition is vital for threat analysis and electronic countermeasures.
• Traditional methods struggle with automatic feature extraction and intra-class compactness.
• Existing closed-set recognition methods are limited by the open nature of space environments.

Purpose of the Study:

• To address limitations in current space radiation source recognition techniques.
• To develop a method effective for both closed-set and open-set recognition scenarios.
• To improve intra-class compactness and inter-class separability in recognition tasks.

Main Methods:

• Proposing a novel multi-scale residual prototype learning network (MSRPLNet).
• Integrating prototype learning principles for enhanced feature representation.
• Designing a joint decision algorithm for open-set recognition of unknown sources.

Main Results:

• Achieved 98.34% accuracy for closed-set recognition of eight Iridium targets.
• Achieved 91.04% accuracy for open-set recognition of eight Iridium targets.
• Demonstrated superior performance compared to existing similar research works.

Conclusions:

• The MSRPLNet is effective for both closed- and open-set space radiation source recognition.
• The proposed method offers significant advantages over traditional and current deep learning approaches.
• The study validates the method's reliability through real-world satellite signal observations.