Model-Based Data Augmentation Applied to Deep Learning Networks for Classification of Micro-Doppler Signatures Using FMCW Radar

Experimental setup adopted for the experimental validation.

 

Summary of "Model-Based Data Augmentation Applied to Deep Learning Networks for Classification of Micro-Doppler Signatures Using FMCW Radar" by Neda Rojhani, Member, IEEE, Marco Passafiume , Member, IEEE, Mehdi Sadeghibakhi, Giovanni Collodi , Member, IEEE, and Alessandro Cidronali , Senior Member, IEEE

Abstract—Deep neural networks (DNNs) have become a relevant subject in the classification of radio frequency signals and remote sensing data. A primary challenge is a tradeoff between obtaining data that are suitable for DNN training and the effort that making experimental measurements requires. Hence, the quality and quantity of data used for the training and testing of models are crucial for effective classifier development. 

 

The training dataset should cover a wide range of cases that synthesize the actual scenarios being classified. This work proposes a novel data augmentation method based on a deterministic model to generate a simulated dataset of radar micro-Doppler signatures suitable for unmanned aerial vehicle (UAV) target classification, without requiring measurement data. It is shown that the DNN trained using the properly generated model-based data offers improved classification accuracy performance. 

 

Results are presented for a two-class classification of the number of UAV motors using a 77-GHz frequency-modulated continuouswave (FMCW) automotive radar system. The effectiveness of
the proposed methodology is proven: a classification accuracy of 78.68% is achieved using a convolutional neural network (CNN) trained using the synthetic dataset, while an accuracy of 66.18%
is achieved by using a typical signal processing data augmentation method on a limited measured dataset.

 

Index Terms—#Dataaugmentation, #deepconvolutionalneuralnetwork (#CNN), #microDoppler signatures, radar classification, remote sensing imagery.

 

https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10004894

 

Model-based dataset generation chain.

 

The paper "Model-Based Data Augmentation Applied to Deep Learning Networks for Classification of Micro-Doppler Signatures Using FMCW Radar" proposes a technique to improve the classification of micro-Doppler signatures using frequency-modulated continuous wave (FMCW) radar by employing model-based data augmentation in conjunction with deep learning networks.

The authors emphasize the significance of accurately classifying micro-Doppler signatures, which provide valuable insights into target motion. However, the scarcity of labeled data makes training deep learning models for this task challenging. To overcome this limitation, the paper introduces model-based data augmentation as a solution.

The proposed method utilizes a mathematical model of the FMCW radar system to generate augmented data. By manipulating various parameters of the model, such as target speed, range, and modulation frequency, a diverse set of synthetic micro-Doppler signatures is synthesized. These augmented samples are combined with the original dataset, creating an expanded training set.

For the classification task, a convolutional neural network (CNN) architecture is employed. The CNN is trained using the augmented dataset, and its performance is evaluated using metrics such as accuracy, precision, recall, and F1 score.

The experimental results demonstrate that the model-based data augmentation approach enhances the classification performance of the CNN. By incorporating augmented samples that capture the variability of real-world scenarios, the CNN learns more robust features and generalizes better to unseen micro-Doppler signatures. A comparison is also made between the performance of the augmented CNN and a baseline CNN trained solely on the original dataset, showing the superiority of the augmented approach.

In summary, the paper introduces a novel technique of model-based data augmentation to enhance the classification of micro-Doppler signatures using FMCW radar. The approach leverages a mathematical model to generate synthetic data, which enriches the training set and enables the CNN to improve its classification capabilities. The experimental results validate the effectiveness of the proposed approach, offering potential advancements in radar-based target classification.