研究目的
To introduce a novel approach for human gait recognition based on backscattered signals from a micro-Doppler radar using deep autoencoder and Bayesian optimization for network design.
研究成果
The proposed technique based on sparse deep autoencoders and Bayesian optimization significantly improves human motion classification compared to existing methods, across three time-frequency and time-scale representations.
研究不足
The study is limited by the specific types of human motions considered and the use of a single radar system. The performance may vary with different motion types or radar systems.
1：Experimental Design and Method Selection:
The study employs three different signal techniques (STFT, S-method, and CWT) for micro-Doppler feature extraction via time-frequency and time-scale representations. A deep autoencoder is used for classification, with its hyperparameters optimized by Bayesian optimization.
2：Sample Selection and Data Sources:
μ-D signals were acquired by a continuous wave Doppler radar from 20 persons performing three types of motions.
3：List of Experimental Equipment and Materials:
ST200 system by RF beam, operating at a carrier frequency of 24 GHz.
4：Experimental Procedures and Operational Workflow:
Signals were downsampled, partitioned into folds, and processed for local patch extraction. The deep autoencoder was trained and evaluated using these patches.
5：Data Analysis Methods:
Performance was measured by classification rate across cross-validation cases.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
