한수지 51(3), 321-327, 2018 Original Article Korean J Fish Aquat Sci 51(3),321-327,2018 확률신경망에의한해저저질의식별 이대재 * 부경대학교해양생산시스템관리학부 Classifying Seafloor Sediments Using a Probabilistic Neural Network Dae-Jae Lee* Division of Marine Production System Management, Pukyong National University, Busan 48513, Korea To classify seafloor sediments using a probabilistic neural network (PNN), the frequency-dependent characteristics of broadband acoustic scattering, which make it possible to qualitatively categorize seabed type, were collected from three different geographical areas in Korea. The echo data samples from three types of seafloor sediment were measured using a chirp sonar system operating over a frequency range of 20-220 khz. The spectrum amplitudes for frequency responses of 35-75 khz were fed into the PNN as input feature parameters. The PNN algorithm could successfully identify three seabed types: mud, mud/shell and concrete sediments. The percentage probabilities of the three seabed types being correctly classified were 86% for mud, 66% for mud/shell and 72% for concrete sediment. Key words: Chirp sonar, Broadband acoustic echoes, Frequency spectrum, Probabilistic neural network, Seafloor sediment classification 서론,,., chirp echo. (Simmons et al., 1996; Saad et al., 2007; Latha et al., 2009; Kuruvilla and Gunavathi, 2014; Lee, 2016). Specht (1990),,, (back-propagation neural network) (probabilistic neural network, PNN). (sample) (probaility density function, PDF) (Rutkowski, 2004; Selekwa et al., 2005). 3 echo (set). chirp echo. chirp 20-220 khz (Lee, 2018), 35-75 khz chirp echo. (sinal to noise ratio, SNR). chirp echo, (mud), (mud/shell),. 재료및방법 해저 echo 신호의수록및해석시스템구성 Lee (2018), chirp https://doi.org/10.5657/kfas.2018.0321 Korean J Fish Aquat Sci 51(3) 321-327, June 2018 This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial Licens (http://creativecommons.org/licenses/by-nc/3.0/) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited. Received 2 May 2018; Revised 16 May 2018; Accepted 16 May 2018 *Corresponding author: Tel: +82. 51. 629. 5889 Fax: +82. 51. 629. 5885 E-mail address: daejael@pknu.ac.kr Copyright 2018 The Korean Society of Fisheries and Aquatic Science 321 pissn:0374-8111, eissn:2287-8815
322 이대재 ( chirp.) (2015 6 23 ) (2015 6 26 ) (5 6 5 m, 2015 7 23 ). echo chirp (Lee, 2018), PC (Dell, Inspirion, USA), 3 chirp (B265LH, Airmar, USA), (DSO model DS-1530, EZ, Korea), (VP-2000, Teledyne Reson, Denmark), USB (P3, 1TB, Samsung, Korea). (G/T 1,737 ) chirp 1.5 m,. 10.5 m, 33.0 m, 1.0 m. chirp 3 chirp (B265LH) 1 1.0 ms, 20-220 khz chirp., 2 (B265LH) (35-75 khz) chirp echo., echo, TVG, A/D (analog to digital) 500 khz,. SNR TVG, 2 echo.,, chirp 40 mm [tungsten carbide sphere with 6% cobalt binder (WC)]., chirp echo, Fig. 2. Fig. 2 2015 6 26 echo. Fig. 2 echo 2,, 2 echo., chirp 0.5 ms, 1 ping 255,, 45 m (60 ms), 0-250 khz. Fig. 2 33.0 m (44 ms) echo., 42-52 ms echo, 255 ping 1. echo. 확률신경망에의한해저저질의식별 (class category) (classifier). 2 (training pattern)., class Fig. 1. Schematic diagram of the chirp data acquisition and processing system for measuring the broadband acoustic echoes from two different seabed sediments (mud, mud/shell) and the concrete sediment of water tank (Lee, 2018). MOSFET, metal oxide semiconductor field effect transistor; BPF, bandpass filter; ADC, analog-to-digital converter; DAC, digital-to-analog converter.
확률신경망에의한해저저질의식별 323 Fig. 2. Frequency-dependent characteristics of acoustic scattering signals acquired at a mud/shell sediment site (Dojangpo bay) of the southern waters, Korea, using the chirp data acquisition and processing system. PDF,., x i class c i PDF,, P (c i x) (1) (Parzen, 1962; Specht, 1990; Selekwa et al., 2005). n 1 i (x-x ji ) T i (x-x j ) P (c i x)= ((2 ) (N 2) N n i ) j=1 2 2, n i i class c i, x, x x=[x 1, x 2, x 3, x 4, x N ] T., x ji i class c i j, (smoothing parameter), N (1)., (1) Fig. 3. Fig. 3,, 4.,, / chirp echo (x),. Fig. 3 ( ) 3 (c i, i=1, 2, 3) (neuron),,, Gaussian (
324 이대재 Fig. 3. Architecture of probabilistic neural network used in this study. ). (1) P (c i x).. 결과및고찰 해저 echo 신호의시간응답특성 (35 07.18 N, 129 03.17 E),, (34 46.04 N, 128 41.09 E),, chirp echo Fig. 4. (PM3D, Marine Electronics Corp., Korea),, /,. Fig. 4 echo 35 m SNR A/D ( 40 mm, WC). Fig. 4 (a) (mud) echo, echo 12.86 ms 16.03 ms Fig. 4. Comparison of the time response characteristics of broadband echo signals recorded from three different sediment types of muddy seabed (a), mud/shell seabed (b) and concrete sediment (c)., 19.04 ms, echo (envelope) (Gaussian)., echo 22.5 ms echo tail. chirp, chirp echo,,., chirp,.,,. Fig. 4 (b) / echo Fig. 4 (a) echo., 43.93 ms 46.20 ms, 52.05 ms, echo., echo (leading edge) echo Fig. 4 (a)
확률신경망에의한해저저질의식별 325 Fig. 5. Comparison of the frequency response characteristics of broadband echo signals recorded from three different sediment types of muddy seabed (black), mud/shell seabed (red) and concrete sediment (blue).., echo (tailing edge) echo., Fig. 4 (c) echo Fig. 4 (a) Fig. 4 (b) / echo., echo 5.71 ms 5.91 ms (peak), 6.16 ms, 6.86 ms.,, echogram. 해저 echo 신호의주파수응답특성, / echo, Fig. 5. Fig. 5,, / echo,, (khz). chirp 20-220 khz chirp, (B265LH) Fig. 6. Frequency spectrum images for 200 broadband echo signals recorded continuously from three different sediment types of muddy seabed (a), mud and shell seabed (b) and concrete sediment (c). echo., Fig. 5 35-75 khz echo. Fig. 5
326 이대재, 3,, 40 khz peak mode 50 khz null mode, 60 khz peak mode. 3 35-75 khz,,., Fig. 3., Fig. 5, /,,, 40 khz, 50 khz 60 khz. 40 khz 50 khz peak null, 60 khz., / peak,, null. echo., chirp, / 200 echo Fig. 6. Fig. 6 ping number, (khz), 0-200 khz. Fig. 6 (a), (b) /., (c). 3, /,, echo null., 40 khz 60 khz 2, 50 khz null., / 38-48 khz 58-72 khz, 50 khz Fig. 7. Bar plots showing the percentage probability of classification for three different sediment types of muddy seabed (a), mud/ shell seabed (b) and concrete sediment (c). The orange, sky blue and green colors in the bar plots indicate muddy seabed, mud/shell seabed and concrete sediment, respectively. null., Fig. 6 (c) 60 khz. 3 echo 35-75 khz,. Fig. 6 150 Fig. 3 class 1 ( ), class 2 ( / ), class 3 ( )., Fig. 6 50
확률신경망에의한해저저질의식별 327. 확률신경망에의한해저저질의식별 Fig. 6 Fig. 3 (pattern layer) 3, Gaussian,, Fig. 3 (summation layer) PDF 3,, Fig. 3 P(c i x), Fig. 7. Fig. 3 Fig. 7, ( ). Fig. 7 (percentage probability),,, bar plot, /. Fig. 7 (a),, /, 3 86%., / 4% 10%., Fig. 7 (b) /,, /, 3 / 66%., / 10% 24%., Fig. 7 (c),, /, 3 72%., / 4% 24%.,, /, 3, 74.7%..,.,,,,,.. 사사 (2017 ). References Kuruvilla J and Gunavathi K. 2014. Lung cancer classification using neural networks for CT images. Comput Methods Programs Biomed 113, 202-209. https://doi.org/10.1016/j. cmpb.2013.10.011. Latha P, Ganesan L and Annadurai S. 2009. Face recognition using neural networks. Sign Pro Inter J 3, 153-160. Lee DJ. 2016. Acoustic identification of six fish species using an artificial neural network. Korean J Fish Aquat Sci 49, 224-233. http://dx.doi.org/10.5657/kfas.2016.0224. Lee DJ. 2018. Performance characteristics of chirp data acquisition and processing system for time-frequency analysis of broadband acoustic scattering signals from fish schools. Korean J Fish Aquat Sci 51, 178-186. http://dx.doi.org/10.5657/ KFAS.2018.0178. Parzen E. 1962. On estimation of a probability density function and mode. Ann Math Statist 33, 1065-1076. Rutkowski L. 2004. Adaptive probabilistic Neural Networks for Pattern classification in time varying environment. IEEE Trans. Neural Networks 15, 811-827. http://dx.doi. org/10.1109/tnn.2004.828757. Saad MHM, Nor MJM, Bustami FRA and Ngadiran R. 2007. Classification of heart abnormalities using artificial neural network. J Appl Sci 7, 820-825. Selekwa MF, Kwigizile V and Mussa RN. 2005. Setting up a probabilistic neural network for classification of highway vehicles. Int J Comput Intell App 5, 411-423. http://dx.doi. org/10.1142/s1469026805001702. Simmons EJ, Armstong F and Copland PJ. 1996. Species identification using wideband backscattering with neural network and discriminant analysis. ICES J Mar Sci 53, 189-195. Specht DF. 1990. Probabilistic neural networks. Neural Networks 3, 109-118.