Recognition of gunshots using artificial neural network

Mohd Faudzi Bin Muhammad, Mohd. Marzuki Mustafa

Research output: Contribution to journalArticle

Abstract

In this paper, a system for automatically identifying guns by their sounds is implemented. The gunshots data are recorded from 3 different types of guns; 155 mm, 105 mm and 90 mm. Power Spectral Densities (PSD) are used to extract the gunshot sound's spectral information. These feature parameters then serve as the input to the artificial neural network (ANN). A three layer feed forward neural network is used with back propagation as the training algorithm. The network size was selected to be 33-8-8-3 (33 input, 8 1 st hidden layer, 8 2 nd hidden layer and 3 output unit). Sample populations of 75 data are used as training data and 75 data serve as testing data. The results show that the trained network can correctly identify gunshot sounds with up to 90% accuracy.

Original languageEnglish
Pages (from-to)58-61
Number of pages4
JournalDefence S and T Technical Bulletin
Volume2
Issue number1
Publication statusPublished - 2009

Fingerprint

Acoustic waves
Neural networks
Feedforward neural networks
Power spectral density
Backpropagation
Testing

Keywords

  • Artificial neural network (ANN)
  • Gunshots
  • Power spectrum densities

ASJC Scopus subject areas

  • Engineering(all)

Cite this

Recognition of gunshots using artificial neural network. / Muhammad, Mohd Faudzi Bin; Mustafa, Mohd. Marzuki.

In: Defence S and T Technical Bulletin, Vol. 2, No. 1, 2009, p. 58-61.

Research output: Contribution to journalArticle

@article{87402235b34440af9d04ca4e3c1ff8eb,
title = "Recognition of gunshots using artificial neural network",
abstract = "In this paper, a system for automatically identifying guns by their sounds is implemented. The gunshots data are recorded from 3 different types of guns; 155 mm, 105 mm and 90 mm. Power Spectral Densities (PSD) are used to extract the gunshot sound's spectral information. These feature parameters then serve as the input to the artificial neural network (ANN). A three layer feed forward neural network is used with back propagation as the training algorithm. The network size was selected to be 33-8-8-3 (33 input, 8 1 st hidden layer, 8 2 nd hidden layer and 3 output unit). Sample populations of 75 data are used as training data and 75 data serve as testing data. The results show that the trained network can correctly identify gunshot sounds with up to 90{\%} accuracy.",
keywords = "Artificial neural network (ANN), Gunshots, Power spectrum densities",
author = "Muhammad, {Mohd Faudzi Bin} and Mustafa, {Mohd. Marzuki}",
year = "2009",
language = "English",
volume = "2",
pages = "58--61",
journal = "Defence S and T Technical Bulletin",
issn = "1985-6571",
publisher = "Science & Technology Research Institute for Defence (STRIDE)",
number = "1",

}

TY - JOUR

T1 - Recognition of gunshots using artificial neural network

AU - Muhammad, Mohd Faudzi Bin

AU - Mustafa, Mohd. Marzuki

PY - 2009

Y1 - 2009

N2 - In this paper, a system for automatically identifying guns by their sounds is implemented. The gunshots data are recorded from 3 different types of guns; 155 mm, 105 mm and 90 mm. Power Spectral Densities (PSD) are used to extract the gunshot sound's spectral information. These feature parameters then serve as the input to the artificial neural network (ANN). A three layer feed forward neural network is used with back propagation as the training algorithm. The network size was selected to be 33-8-8-3 (33 input, 8 1 st hidden layer, 8 2 nd hidden layer and 3 output unit). Sample populations of 75 data are used as training data and 75 data serve as testing data. The results show that the trained network can correctly identify gunshot sounds with up to 90% accuracy.

AB - In this paper, a system for automatically identifying guns by their sounds is implemented. The gunshots data are recorded from 3 different types of guns; 155 mm, 105 mm and 90 mm. Power Spectral Densities (PSD) are used to extract the gunshot sound's spectral information. These feature parameters then serve as the input to the artificial neural network (ANN). A three layer feed forward neural network is used with back propagation as the training algorithm. The network size was selected to be 33-8-8-3 (33 input, 8 1 st hidden layer, 8 2 nd hidden layer and 3 output unit). Sample populations of 75 data are used as training data and 75 data serve as testing data. The results show that the trained network can correctly identify gunshot sounds with up to 90% accuracy.

KW - Artificial neural network (ANN)

KW - Gunshots

KW - Power spectrum densities

UR - http://www.scopus.com/inward/record.url?scp=84860126232&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84860126232&partnerID=8YFLogxK

M3 - Article

AN - SCOPUS:84860126232

VL - 2

SP - 58

EP - 61

JO - Defence S and T Technical Bulletin

JF - Defence S and T Technical Bulletin

SN - 1985-6571

IS - 1

ER -