Generation of Human Micro-Doppler Signature Based on Layer-Reduced Deep Convolutional Generative Adversarial Network

Area of Science:

• Radar signal processing
• Machine learning for human activity recognition

Background:

• Human activity recognition (HAR) using radar micro-Doppler is a growing research area.
• Radar offers practical advantages for HAR, but classifier performance relies on extensive datasets.
• Existing radar micro-Doppler databases are often limited in size due to creation challenges.

Purpose of the Study:

• To propose a novel method for generating radar micro-Doppler data of the human body.
• To address the limitations of small datasets in radar-based HAR.

Main Methods:

• Utilized a deep convolutional generative adversarial network (DCGAN) for data generation.
• Converted existing motion databases into simulated model-based radar data as input for the DCGAN.
• Employed a generative approach to synthesize radar micro-Doppler signatures.

Main Results:

• Successfully generated radar micro-Doppler data for human activities.
• Demonstrated the effectiveness of the DCGAN method even with limited input data.
• Validated the proposed method through simulation.

Conclusions:

• The proposed DCGAN-based method offers a viable solution for augmenting limited radar micro-Doppler datasets.
• This approach can significantly aid the development of more robust HAR systems.
• The technique shows promise for advancing radar-based activity recognition research.