Micro-Doppler feature extraction using convolutional auto-encoders for low latency target classification

Parashar, Karthick; Oveneke, Meshia Cédric; Rykunov, Maxim; Sahli, Hichem; Bourdoux, André

doi:10.1109/radar.2017.7944488

Cited by 24 publications

(17 citation statements)

References 16 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…More sophisticated DCNN architectures were used later, including 7-layer DCNN [13], transfer learned AlexNet and VGG-16 network [14] and a three-layer semisupervised auto-encoder [15]. New problems such as low latency classification [16] and multi-target human gait classification [9,17] have also been taken into consideration.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Personnel Recognition and Gait Classification Based on Multistatic Micro-Doppler Signatures Using Deep Convolutional Neural Networks

Chen

Fioranelli

et al. 2018

IEEE Geosci. Remote Sensing Lett.

140

View full text Add to dashboard Cite

Personnel recognition and gait classification based on multistatic micro-doppler signatures using deep convolutional neural networks. IEEE Geoscience and Remote Sensing Letters, (doi:10.1109/LGRS.2018.2806940) This is the author's final accepted version.There may be differences between this version and the published version. You are advised to consult the publisher's version if you wish to cite from it.http://eprints.gla.ac.uk/156827/ Abstract-In this letter we propose two methods for personnel recognition and gait classification using deep convolutional neural networks (DCNNs) based on multistatic radar micro-Doppler signatures. Previous DCNN based schemes have mainly focused on monostatic scenarios, whereas directional diversity offered by multistatic radar is exploited in our work to improve classification accuracy. We first propose the voted monostatic DCNN method (VMo-DCNN), which trains DCNNs on each receiver node separately, and fuses the results by binary voting. By merging the fusion step into the network architecture, we further propose the multistatic DCNN method (Mul-DCNN), which performs slightly better than VMo-DCNN. These methods are validated on real data measured with a 2.4 GHz multistatic radar system. Experimental results show that Mul-DCNN achieves over 99% accuracy in armed/unarmed gait classification using only 20% training data and similar performance in two-class personnel recognition using 50% training data, which are higher than the accuracy obtained by performing DCNN on a single radar node.

show abstract

Section: Introductionmentioning

confidence: 99%

“…A novel work [16] improve classification accuracy, using deep learning and boosting trees. However, the data are measured by monostatic radar at different aspect angles rather than by a multistatic radar system simultaneously.…”

Section: Introductionmentioning

confidence: 99%

Personnel Recognition and Gait Classification Based on Multistatic Micro-Doppler Signatures Using Deep Convolutional Neural Networks

Chen

Fioranelli

et al. 2018

IEEE Geosci. Remote Sensing Lett.

140

View full text Add to dashboard Cite

show abstract

“…Three different neural network architectures have been exploreda downscaled version of the network VGG16, utilizing the same block structure; the very deep ResNET-50 [31], which uses shortcuts between network blocks to avoid overfitting and achieve better generalization; and an innovative CNN+LSTM (Long Short-Term Memory) architecture, which is able to extract features from micro-Doppler spectrogram segments, and learn their representation as time series (sequences of data). This is an innovative approach, as the radar data will be considered by the LSTM network part not as snapshot spectrograms images (as currently done in many works in the literature [22][23][24][25][26][27][28]), but as temporal data sequences. Although demonstrated on preliminary results on a small experimental dataset, this classification approach may prove well suited to radar data, exploiting the inherent information from a sequence of radar waveforms, rather than casting the problem as classification of images.…”

Section: Introductionmentioning

confidence: 99%

Practical classification of different moving targets using automotive radar and deep neural networks

Angelov

Robertson

Murray-Smith

et al. 2018

IET Radar, Sonar & Navigation

109

View full text Add to dashboard Cite

In this work, we present results for classification of different classes of targets (car, single and multiple people, bicycle) using automotive radar data and different neural networks. A fast implementation of radar algorithms for detection, tracking, and micro-Doppler extraction is proposed in conjunction with the automotive radar transceiver TEF810X and microcontroller unit SR32R274 manufactured by NXP Semiconductors. Three different types of neural networks are considered, namely a classic convolutional network, a residual network, and a combination of convolutional and recurrent network, for different classification problems across the 4 classes of targets recorded. Considerable accuracy (close to 100% in some cases) and low latency of the radar pre-processing prior to classification (approximately 0.55s to produce a 0.5s long spectrogram) are demonstrated in this paper, and possible shortcomings and outstanding issues are discussed.

show abstract

“…The effect of different dwell times was also briefly observed. Convolutional neural networks were also exploited in (Parashar et al, 2017) for identification of target classes in the context of automotive radar, for example pedestrians and bicycles vs cars and vehicles, and in (Seyfioglu, 2017) to discriminate different human actions and activities in the context of ambient assisted living and remote health monitoring. In both cases, the micro-Doppler signatures were used directly as inputs to the network.…”

Section: Introductionmentioning

confidence: 99%

Multistatic radar classification of armed vs unarmed personnel using neural networks

et al. 2017

View full text Add to dashboard Cite

This paper investigates an implementation of an array of distributed neural networks, operating together to classify between unarmed and potentially armed personnel in areas under surveillance using ground based radar. Experimental data collected by the University College London (UCL) multistatic radar system NetRAD is analysed. Neural networks were introduced to the extracted micro-Doppler data in order to classify between the two scenarios, and accuracy above 98% is demonstrated on the validation data, showing an improvement over methodologies based on classifiers, where human intervention is required. The main advantage of using neural networks is to bypass the manual extraction process of handcrafted features from the radar data, where thresholds and parameters need to be tuned by human operators. Different network architectures are explored, from feed-forward networks to stacked auto-encoders, with the advantages of deep topologies being capable of classifying the spectrograms (Doppler-time patterns) directly. Significant parameters concerning the actual deployment of the networks are also investigated, for example the dwell time (i.e. how long the radar needs to stare at a target in order to achieve classification), and the robustness of the networks in classifying data from new people, whose signatures were unseen during the training stage. Finally, a data ensembling technique is also presented which utilises a weighted decision approach, established beforehand, utilising information from all three sensors, and yielding stable classification accuracies of 99% or more, across all monitored zones.

show abstract

Micro-Doppler feature extraction using convolutional auto-encoders for low latency target classification

Cited by 24 publications

References 16 publications

Personnel Recognition and Gait Classification Based on Multistatic Micro-Doppler Signatures Using Deep Convolutional Neural Networks

Personnel Recognition and Gait Classification Based on Multistatic Micro-Doppler Signatures Using Deep Convolutional Neural Networks

Practical classification of different moving targets using automotive radar and deep neural networks

Multistatic radar classification of armed vs unarmed personnel using neural networks

Contact Info

Product

Resources

About