DopNet: A Deep Convolution Neural Network to Recognize Armed and Unarmed Human Targets

Qingchao Chen, Yang Liu, Francesco Fioranelli, Matthiew Ritchie, Bo Tan, Kevin Chetty

Research output: Contribution to journalArticle

Abstract

Abstract—The work presented in this paper aims to distinguish
between armed or unarmed personnel using multi-static radar
data and advanced Doppler processing. We propose two modified Deep Convolutional Neural Networks (DCNN) termed SCDopNet and MC-DopNet for mono-static and multi-static microDoppler signature (µ-DS) classification. Differentiating armed
and unarmed walking personnel is challenging due to the effect
of aspect angle and channel diversity in real-world scenarios.
In addition, DCNN easily overfits the relatively small-scale µ-DS
dataset. To address these problems, the work carried out in this
paper makes three key contributions: first, two effective schemes
including data augmentation operation and a regularization
term are proposed to train SC-DopNet from scratch. Next,
a factor analysis of the SC-DopNet are conducted based on
various operating parameters in both the processing and radar
operations. Thirdly, to solve the problem of aspect angle diversity
for µ-DS classification, we design MC-DopNet for multi-static µDS which is embedded with two new fusion schemes termed
as Greedy Importance Reweighting (GIR) and `21-Norm. These
two schemes are based on two different strategies and have been
evaluated experimentally: GIR uses a “win by sacrificing worst
case” whilst `21-Norm adopts a “win by sacrificing best case”
approach. The SC-DopNet outperforms the non-deep methods
by 12.5% in average and the proposed MC-DopNet with two
fusion methods outperforms the conventional binary voting by
1.2% in average. Note that we also argue and discuss how to
utilize the statistics of SC-DopNet results to infer the selection
of fusion strategies for MC-DopNet under different experimental
scenarios.
LanguageEnglish
Pages(In-Press)
JournalIEEE Sensors Journal
Volume(In-Press)
StateAccepted/In press - 10 Jan 2019

Fingerprint

Convolution
convolution integrals
Fusion reactions
fusion
personnel
Neural networks
norms
Radar
Personnel
multistatic radar
voting
factor analysis
walking
radar data
Factor analysis
Processing
radar
signatures
Statistics
statistics

Cite this

Chen, Q., Liu, Y., Fioranelli, F., Ritchie, M., Tan, B., & Chetty, K. (Accepted/In press). DopNet: A Deep Convolution Neural Network to Recognize Armed and Unarmed Human Targets. IEEE Sensors Journal, (In-Press), (In-Press).

DopNet : A Deep Convolution Neural Network to Recognize Armed and Unarmed Human Targets. / Chen, Qingchao; Liu, Yang; Fioranelli, Francesco; Ritchie, Matthiew; Tan, Bo; Chetty, Kevin.

In: IEEE Sensors Journal, Vol. (In-Press), 10.01.2019, p. (In-Press).

Research output: Contribution to journalArticle

Chen, Q, Liu, Y, Fioranelli, F, Ritchie, M, Tan, B & Chetty, K 2019, 'DopNet: A Deep Convolution Neural Network to Recognize Armed and Unarmed Human Targets' IEEE Sensors Journal, vol. (In-Press), pp. (In-Press).
Chen Q, Liu Y, Fioranelli F, Ritchie M, Tan B, Chetty K. DopNet: A Deep Convolution Neural Network to Recognize Armed and Unarmed Human Targets. IEEE Sensors Journal. 2019 Jan 10;(In-Press):(In-Press).
Chen, Qingchao ; Liu, Yang ; Fioranelli, Francesco ; Ritchie, Matthiew ; Tan, Bo ; Chetty, Kevin. / DopNet : A Deep Convolution Neural Network to Recognize Armed and Unarmed Human Targets. In: IEEE Sensors Journal. 2019 ; Vol. (In-Press). pp. (In-Press)
@article{166ed4b9fd4c423fa60173d2355a6a0f,
title = "DopNet: A Deep Convolution Neural Network to Recognize Armed and Unarmed Human Targets",
abstract = "Abstract—The work presented in this paper aims to distinguishbetween armed or unarmed personnel using multi-static radardata and advanced Doppler processing. We propose two modified Deep Convolutional Neural Networks (DCNN) termed SCDopNet and MC-DopNet for mono-static and multi-static microDoppler signature (µ-DS) classification. Differentiating armedand unarmed walking personnel is challenging due to the effectof aspect angle and channel diversity in real-world scenarios.In addition, DCNN easily overfits the relatively small-scale µ-DSdataset. To address these problems, the work carried out in thispaper makes three key contributions: first, two effective schemesincluding data augmentation operation and a regularizationterm are proposed to train SC-DopNet from scratch. Next,a factor analysis of the SC-DopNet are conducted based onvarious operating parameters in both the processing and radaroperations. Thirdly, to solve the problem of aspect angle diversityfor µ-DS classification, we design MC-DopNet for multi-static µDS which is embedded with two new fusion schemes termedas Greedy Importance Reweighting (GIR) and `21-Norm. Thesetwo schemes are based on two different strategies and have beenevaluated experimentally: GIR uses a “win by sacrificing worstcase” whilst `21-Norm adopts a “win by sacrificing best case”approach. The SC-DopNet outperforms the non-deep methodsby 12.5{\%} in average and the proposed MC-DopNet with twofusion methods outperforms the conventional binary voting by1.2{\%} in average. Note that we also argue and discuss how toutilize the statistics of SC-DopNet results to infer the selectionof fusion strategies for MC-DopNet under different experimentalscenarios.",
author = "Qingchao Chen and Yang Liu and Francesco Fioranelli and Matthiew Ritchie and Bo Tan and Kevin Chetty",
year = "2019",
month = "1",
day = "10",
language = "English",
volume = "(In-Press)",
pages = "(In--Press)",
journal = "IEEE Sensors Journal",
issn = "1530-437X",
publisher = "Institute of Electrical and Electronics Engineers",

}

TY - JOUR

T1 - DopNet

T2 - IEEE Sensors Journal

AU - Chen,Qingchao

AU - Liu,Yang

AU - Fioranelli,Francesco

AU - Ritchie,Matthiew

AU - Tan,Bo

AU - Chetty,Kevin

PY - 2019/1/10

Y1 - 2019/1/10

N2 - Abstract—The work presented in this paper aims to distinguishbetween armed or unarmed personnel using multi-static radardata and advanced Doppler processing. We propose two modified Deep Convolutional Neural Networks (DCNN) termed SCDopNet and MC-DopNet for mono-static and multi-static microDoppler signature (µ-DS) classification. Differentiating armedand unarmed walking personnel is challenging due to the effectof aspect angle and channel diversity in real-world scenarios.In addition, DCNN easily overfits the relatively small-scale µ-DSdataset. To address these problems, the work carried out in thispaper makes three key contributions: first, two effective schemesincluding data augmentation operation and a regularizationterm are proposed to train SC-DopNet from scratch. Next,a factor analysis of the SC-DopNet are conducted based onvarious operating parameters in both the processing and radaroperations. Thirdly, to solve the problem of aspect angle diversityfor µ-DS classification, we design MC-DopNet for multi-static µDS which is embedded with two new fusion schemes termedas Greedy Importance Reweighting (GIR) and `21-Norm. Thesetwo schemes are based on two different strategies and have beenevaluated experimentally: GIR uses a “win by sacrificing worstcase” whilst `21-Norm adopts a “win by sacrificing best case”approach. The SC-DopNet outperforms the non-deep methodsby 12.5% in average and the proposed MC-DopNet with twofusion methods outperforms the conventional binary voting by1.2% in average. Note that we also argue and discuss how toutilize the statistics of SC-DopNet results to infer the selectionof fusion strategies for MC-DopNet under different experimentalscenarios.

AB - Abstract—The work presented in this paper aims to distinguishbetween armed or unarmed personnel using multi-static radardata and advanced Doppler processing. We propose two modified Deep Convolutional Neural Networks (DCNN) termed SCDopNet and MC-DopNet for mono-static and multi-static microDoppler signature (µ-DS) classification. Differentiating armedand unarmed walking personnel is challenging due to the effectof aspect angle and channel diversity in real-world scenarios.In addition, DCNN easily overfits the relatively small-scale µ-DSdataset. To address these problems, the work carried out in thispaper makes three key contributions: first, two effective schemesincluding data augmentation operation and a regularizationterm are proposed to train SC-DopNet from scratch. Next,a factor analysis of the SC-DopNet are conducted based onvarious operating parameters in both the processing and radaroperations. Thirdly, to solve the problem of aspect angle diversityfor µ-DS classification, we design MC-DopNet for multi-static µDS which is embedded with two new fusion schemes termedas Greedy Importance Reweighting (GIR) and `21-Norm. Thesetwo schemes are based on two different strategies and have beenevaluated experimentally: GIR uses a “win by sacrificing worstcase” whilst `21-Norm adopts a “win by sacrificing best case”approach. The SC-DopNet outperforms the non-deep methodsby 12.5% in average and the proposed MC-DopNet with twofusion methods outperforms the conventional binary voting by1.2% in average. Note that we also argue and discuss how toutilize the statistics of SC-DopNet results to infer the selectionof fusion strategies for MC-DopNet under different experimentalscenarios.

M3 - Article

VL - (In-Press)

SP - (In-Press)

JO - IEEE Sensors Journal

JF - IEEE Sensors Journal

SN - 1530-437X

ER -