Change Point Detection for Fine-Grained MFR Work Modes with Multi-Head Attention-Based Bi-LSTM Network

Area of Science:

• Radar Systems Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Detecting Multi-Functional Radar (MFR) work mode changes is crucial for Electronic Support Measure (ESM) systems.
• Challenges include unknown segment durations and complex, fine-grained MFR modes difficult for traditional methods.

Purpose of the Study:

• To propose a deep learning framework for accurate fine-grained change point detection (CPD) in MFR work modes.
• To address the limitations of existing statistical and basic learning models in complex radar environments.

Main Methods:

• Established a fine-grained MFR work mode model.
• Utilized a multi-head attention-based bi-directional long short-term memory network to analyze pulse sequences.
• Incorporated temporal features and optimized training with improved label configuration and loss functions to address label sparsity.

Main Results:

• The proposed framework demonstrated improved parameter-level CPD performance compared to existing methods.
• Achieved a 4.15% increase in F1-score under hybrid non-ideal conditions.
• Effectively mitigated the label sparsity problem during training.

Conclusions:

• The deep learning framework offers a robust solution for fine-grained MFR work mode CPD.
• Enhances the situational awareness capabilities of Electronic Support Measure (ESM) systems.
• Represents a significant advancement over traditional CPD techniques in complex radar scenarios.