Mobile QR Code QR CODE : The Transactions P of the Korean Institute of Electrical Engineers

  1. (Dept. of Electronics Engineering, Interdisciplinary Program in IT-Bio Convergence System, Chosun University, Korea.)



person identification, electrocardiogram, convolutional neural networks, time-frequency transform, Short-Time Fourier Transform, Fourier Synchrosqueezed Transform

1. ์„œ ๋ก 

์ธ๊ณต์ง€๋Šฅ์˜ ํ•ต์‹ฌ ๊ธฐ์ˆ ์ธ ๋”ฅ๋Ÿฌ๋‹์œผ๋กœ ์ธํ•ด ์ธ๊ณต์ง€๋Šฅ์˜ ๊ธฐ์ˆ  ๋ถ„์•ผ๋Š” ๋†์—…, ์˜๋ฃŒ, ๊ธˆ์œต, ๋ณด์•ˆ ๋“ฑ์œผ๋กœ ํ™•๋Œ€๋˜๊ณ  ์ธ๊ฐ„์˜ ์‚ถ๊ณผ ์ผ์˜ ์งˆ์ด ํ–ฅ์ƒ๋˜๊ณ  ์žˆ๋‹ค. ํ•˜์ง€๋งŒ ๊ธฐ์ˆ ์˜ ํ™•์‚ฐ๊ณผ ๋ฐœ์ „์œผ๋กœ ์‚ฌ์ด๋ฒ„ ๋ณด์•ˆ์˜ ๊ณต๊ฒฉ์ด ์ฆ๊ฐ€ํ•˜๊ณ  ์žˆ๋‹ค[1]. ์ด๋Ÿฌํ•œ ๊ณต๊ฒฉ์„ ๋ง‰๊ธฐ ์œ„ํ•ด ๋‹ค์–‘ํ•œ ๊ฐœ์ธ์ธ์‹ ๊ธฐ์ˆ ์ด ์—ฐ๊ตฌ๋˜๊ณ  ์žˆ๋‹ค.

๊ธฐ์กด์˜ ๊ฐœ์ธ์ธ์‹ ๋ฐฉ์‹์€ ํŒจ์Šค์›Œ๋“œ, ๊ณต์ธ์ธ์ฆ์„œ, OTP ๋“ฑ์„ ์‚ฌ์šฉํ•˜์—ฌ ๊ฐœ์ธ์˜ ์ •๋ณด๋ฅผ ๋ณดํ˜ธํ–ˆ์ง€๋งŒ, ๋ถ„์‹ค, ๋„๋‚œ ๋˜๋Š” ๋ชจ๋ฐฉ ๋“ฑ์˜ ์œ„ํ—˜์œผ๋กœ ์‹ ๋ขฐํ•˜๊ธฐ ์–ด๋ ต๊ธฐ ๋•Œ๋ฌธ์— ๊ฐœ์ธ์˜ ์ƒ์ฒด ์ •๋ณด๋ฅผ ์ด์šฉํ•œ ๊ฐœ์ธ์ธ์‹ ๊ธฐ์ˆ ์ด ์—ฐ๊ตฌ๋˜๊ณ  ์žˆ๋‹ค[2][3]. ์ƒ์ฒด์ธ์‹์€ ์‹ ์ฒด์ ์ธ ํŠน์„ฑ๊ณผ ํ–‰๋™ํ•™์  ํŠน์„ฑ์œผ๋กœ ๋‚˜๋ˆŒ ์ˆ˜ ์žˆ๋‹ค. ์‹ ์ฒด์ ์ธ ํŠน์„ฑ์€ ์–ผ๊ตด, ์ง€๋ฌธ, ํ™์ฑ„ ๋“ฑ์ด ์žˆ๊ณ , ํ–‰๋™ํ•™์  ํŠน์„ฑ์€ ์‚ฌ๋žŒ์˜ ํ–‰๋™์  ํŠน์ง•์„ ์˜๋ฏธํ•˜๋ฉฐ ๊ฑธ์Œ๊ฑธ์ด, ์Œ์„ฑ, ์„œ๋ช… ๋˜๋Š” ํ•„์ฒด ๋“ฑ์œผ๋กœ ์ด๋Ÿฌํ•œ ํŠน์„ฑ๋“ค์€ ๋น ๋ฅด๊ฒŒ ์ธ์‹์ด ๊ฐ€๋Šฅํ•˜๊ธฐ ๋•Œ๋ฌธ์— ํ˜„์žฌ ๋งŽ์ด ์‚ฌ์šฉ๋˜๊ณ  ์žˆ๋Š” ์ธ์‹ ๋ฐฉ๋ฒ•์ด๋‹ค. ํ•˜์ง€๋งŒ ์‹ ์ฒด์ ์ธ ํŠน์„ฑ๊ณผ ํ–‰๋™ํ•™์  ํŠน์„ฑ์˜ ๊ฒฝ์šฐ ๋‹ค๋ฅธ ์‚ฌ๋žŒ๋“ค๊ณผ ์†Œํ†ตํ•˜๋ฉฐ ์ •๋ณด๋ฅผ ๊ณต์œ ํ•˜๋Š” ์†Œ์…œ ๋„คํŠธ์›Œํฌ์—์„œ ๊ฐœ์ธ์˜ ์ •๋ณด๋Š” ์‹ ์ฒด ์™ธ๋ถ€๋กœ ์‰ฝ๊ฒŒ ๋ณด์—ฌ์ง€๊ธฐ ๋•Œ๋ฌธ์— ์‚ฌ์ง„ ๋˜๋Š” ๊ธฐ๋ก์œผ๋กœ ๋ˆ„๊ตฌ๋‚˜ ๋ณผ ์ˆ˜ ์žˆ๋Š” ์ •๋ณด๋กœ ์œ„๋ณ€์กฐ ์œ„ํ—˜, ๋„์šฉ, ์ฃผ๋ณ€ ํ™˜๊ฒฝ์˜ ์˜ํ–ฅ์„ ๋ฏธ์น˜๋Š” ๋ฌธ์ œ์ ์ด ์žˆ๋‹ค. ์ด๋Ÿฌํ•œ ๋ฌธ์ œ์ ์„ ํ•ด๊ฒฐํ•˜๊ธฐ ์œ„ํ•ด ์™ธ๋ถ€๋กœ ๋ณด์—ฌ์ง€๋Š” ์ƒ์ฒด์ธ์‹์ด ์•„๋‹Œ ์‹ ์ฒด ๋‚ด๋ถ€์˜ ์‹ ํ˜ธ์ธ ๊ทผ์ „๋„, ๋‡Œ์ „๋„, ์‹ฌ์ „๋„์™€ ๊ฐ™์€ ์‹ ํ˜ธ๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ๊ฐœ์ธ์„ ์ธ์‹ํ•˜๋Š” ์—ฐ๊ตฌ๊ฐ€ ํ™œ๋ฐœํžˆ ์ง„ํ–‰๋˜๊ณ  ์žˆ๋‹ค.

๊ทผ์ „๋„๋Š” ๊ทผ์œก์˜ ์ˆ˜์ถ•์— ์˜ํ•ด ๋ฐœ์ƒํ•œ ์‹ ํ˜ธ๋ฅผ ๊ธฐ๋กํ•œ ๊ฒƒ์œผ๋กœ, ์˜๋ฃŒ๋ถ„์•ผ, ๋ชจ์…˜์ธ์‹ ๋“ฑ์œผ๋กœ ํ™œ์šฉ๋˜๊ณ  ์žˆ๋‹ค. ํ•˜์ง€๋งŒ ๊ทผ์ „๋„๋ฅผ ์ธ์‹ํ•˜๊ธฐ ์œ„ํ•ด์„œ๋Š” ๊ฐ ๋ชจ์…˜์˜ ์‹ ํ˜ธ๋ฅผ ์„ค์ •ํ•ด์•ผํ•˜๊ณ  ์‹ ํ˜ธ ๋ถ„์„์„ ์œ„ํ•ด ์ •ํ™•ํ•œ ์œ„์น˜์— ๋ถ€์ฐฉํ•ด์•ผ ํ•˜๋Š” ๋‹จ์ ์ด ์žˆ๋‹ค. ๋‡Œ์ „๋„๋Š” ๋‡Œ์˜ ์ „๊ธฐ์ ์ธ ์‹ ํ˜ธ๋ฅผ ์ „๊ทน์„ ํ†ตํ•ด ๊ธฐ๋กํ•˜๋Š” ๊ฒƒ์œผ๋กœ, ๊ฐ์„ฑ ICT[4], ๋‰ด๋กœ๋งˆ์ผ€ํŒ…, ๋‡Œ-์ปดํ“จํ„ฐ ์ธํ„ฐํŽ˜์ด์Šค[5], ์˜๋ฃŒ๋ถ„์•ผ ๋“ฑ์—์„œ ํ™œ์šฉ๋˜๊ณ  ์žˆ๋‹ค. ํ•˜์ง€๋งŒ ๋‹ค๋ฅธ ์ธ์‹ ๋ฐฉ๋ฒ•๋“ค๊ณผ ๋‹ค๋ฅด๊ฒŒ ๋‡Œ์ „๋„๋Š” ์„ผ์„œ๋ฅผ ๋จธ๋ฆฌ์— ์ฐฉ์šฉํ•˜์—ฌ ์ธก์ •ํ•˜๊ธฐ ๋•Œ๋ฌธ์— ๊ฑฐ๋ถ€๊ฐ์ด ์žˆ๊ณ , ์›ํ•˜๋Š” ์ธก์ • ์œ„์น˜๋ฅผ ์ •ํ™•ํ•˜๊ฒŒ ๋ถ€์ฐฉ์„ ํ•ด์•ผ ํ•˜๋ฉฐ, ๋‘๊ฐœ๊ณจ์„ ๊ฑฐ์น˜๋ฉด์„œ ์‹ ํ˜ธ๊ฐ€ ์™œ๊ณก๋˜๋Š” ๋ฌธ์ œ์ ์ด ์žˆ๋‹ค. ์‹ฌ์ „๋„๋Š” ์‹ฌ๋ฐ•๋™์˜ ์ „๊ธฐ์ ์ธ ์‹ ํ˜ธ๋ฅผ ์ „๋ฅ˜์— ์˜ํ•ด ๊ธฐ๋กํ•œ ๊ฒƒ์œผ๋กœ, ์ •์ƒ์ ์ธ ์‚ฌ๋žŒ์˜ ์‹ฌ์ „๋„ ์‹ ํ˜ธ๋Š” PํŒŒ, QRS ๋ณตํ•ฉํŒŒ, TํŒŒ๋กœ ๊ตฌ์„ฑ๋˜์–ด ์žˆ์œผ๋ฉฐ, ์ฃผ๊ธฐ์ ์œผ๋กœ ๋‚˜ํƒ€๋‚œ๋‹ค. PํŒŒ๋Š” ์‹ฌ๋ฐฉ์˜ ํƒˆ๋ถ„๊ทน, QRS ๋ณตํ•ฉํŒŒ๋Š” ์‹ฌ์‹ค์˜ ํƒˆ๋ถ„๊ทน, TํŒŒ๋Š” ์‹ฌ์‹ค์˜ ์žฌ๋ถ„๊ทน์— ์˜ํ•ด ํŒŒํ˜•์ด ๋‚˜ํƒ€๋‚œ๋‹ค. ์‹ฌ์ „๋„๋Š” ๊ฐœ์ธ๋งˆ๋‹ค ์„ฑ๋ณ„, ๋‚˜์ด, ์‹ฌ์žฅ์˜ ์œ„์น˜, ํฌ๊ธฐ์— ๋”ฐ๋ผ ๋‹ค๋ฅด๊ธฐ ๋•Œ๋ฌธ์— ๊ฐœ์ธ์˜ ๊ณ ์œ ํ•œ ์‹ ํ˜ธ์ด๋‹ค. ๊ทผ์ „๋„์™€ ๋‡Œ์ „๋„์™€ ๋‹ค๋ฅด๊ฒŒ ์‹ฌ์ „๋„์˜ ์ธก์ • ๋ฐฉ๋ฒ•์€ ์–‘์ชฝ ์†๋ชฉ์—์„œ ๊ฐ„ํŽธํ•˜๊ฒŒ ์‹ ํ˜ธ๋ฅผ ์ธก์ •ํ•  ์ˆ˜ ์žˆ๊ณ , ๊ฐœ์ธ์˜ ๊ณ ์œ ํ•œ ์‹ ํ˜ธ๋กœ ์œ„๋ณ€์กฐ๊ฐ€ ์–ด๋ ค์šฐ๋ฉฐ, ์งˆ๋ณ‘๊ณผ ์ƒ๊ด€์—†์ด ๋ชจ๋“  ์‚ฌ๋žŒ์ด ์ง€๋‹ˆ๊ณ  ์žˆ๋Š” ์‹ ํ˜ธ์ด๋‹ค[6]. ์‹ฌ์ „๋„๋Š” ์ƒ์ฒด์ธ์‹์„ ์œ„ํ•œ ๊ธฐ์ˆ ์˜ 7๊ฐ€์ง€ ํŠน์„ฑ์ธ ๋ณดํŽธ์„ฑ, ์˜๊ตฌ์„ฑ, ์œ ์ผ์„ฑ, ํš๋“์„ฑ, ์ •ํ™•์„ฑ, ๊ธฐ๋งŒ์„ฑ, ์ ‘๊ทผ์„ฑ์„ ํฌํ•จํ•œ๋‹ค[7]. ๋˜ํ•œ ๋ณด์•ˆ์„ฑ์ด ๊ฐ€์žฅ ์šฐ์ˆ˜ํ•˜๊ธฐ ๋•Œ๋ฌธ์— ์‹ฌ์ „๋„๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ๊ฐœ์ธ์‹๋ณ„์„ ํ•œ๋‹ค.

์‹ฌ์ „๋„์˜ ํŠน์ง• ์ถ”์ถœ์„ ์œ„ํ•ด Handcrafted์™€ Non-Handcrafted์˜ ํŠน์ง•๋ฐฉ๋ฒ•์„ ์‚ฌ์šฉํ•  ์ˆ˜ ์žˆ๋‹ค. Handcrafted ํŠน์ง•๋ฐฉ๋ฒ•์€ ์‹ฌ์ „๋„ ์‹ ํ˜ธ์˜ ํ”ผํฌ, ์ง„ํญ, ์‹œ๊ฐ„ ๊ฐ„๊ฒฉ ๋“ฑ์„ ์‚ฌ๋žŒ์ด ์ง์ ‘ ์ถ”์ถœํ•˜๋Š” ๋ฐฉ๋ฒ•์œผ๋กœ ์‹ ํ˜ธ ์™œ๊ณก๊ณผ ์„ฑ๋Šฅ ์ €ํ•˜ ๋ฌธ์ œ๊ฐ€ ๋ฐœ์ƒํ•œ๋‹ค[8][9]. Non-Handcrafted ํŠน์ง•๋ฐฉ๋ฒ•์€ LSTM๊ณผ 2D-CNN์„ ์‚ฌ์šฉํ•˜๋ฉฐ, ํ•™์Šต์„ ํ†ตํ•ด ํŠน์ง•์„ ์ถ”์ถœํ•˜๋Š” ๋ฐฉ๋ฒ•์œผ๋กœ ์‚ฌ๋žŒ์ด ์ง์ ‘ ์ถ”์ถœํ•  ํ•„์š” ์—†๋‹ค. ๋˜ํ•œ ์‹ ํ˜ธ์˜ ํ˜•ํƒœ๋ฅผ ๊ธฐ๋ฐ˜์œผ๋กœ ํŠน์ง•์„ ์ถ”์ถœํ•˜๊ธฐ ๋•Œ๋ฌธ์— R-ํ”ผํฌ๋ฅผ ๊ฒ€์ถœํ•˜์—ฌ ์‹ ํ˜ธ ๋ถ„ํ•  ๊ณผ์ •์ด ํ•„์š”ํ•˜๋‹ค.

๋ณธ ๋…ผ๋ฌธ์€ ๊ฐœ๊ฐœ์ธ๋งˆ๋‹ค ๋‹ค๋ฅด๊ฒŒ ๋‚˜ํƒ€๋‚˜๋Š” ์‹ฌ์ „๋„ ์‹ ํ˜ธ๋ฅผ ์ด์šฉํ•ด ๊ฐœ์ธ์‹๋ณ„ ํ•˜๋Š” ๊ฒƒ์œผ๋กœ ์‹ฌ์ „๋„ ๋ฒค์น˜๋งˆํ‚น ๋ฐ์ดํ„ฐ๋ฒ ์ด์Šค์ธ PTB Database[10]๋ฅผ ์‚ฌ์šฉํ•˜๊ณ , Non-Handcrafted ํŠน์ง•๋ฐฉ๋ฒ•์„ ์œ„ํ•ด R-ํ”ผํฌ๋ฅผ ๊ฒ€์ถœํ•˜์—ฌ ์ผ์ •ํ•œ ๊ฐ„๊ฒฉ์œผ๋กœ ์‹ ํ˜ธ๋ฅผ ๋ถ„ํ• ํ•˜์˜€๋‹ค. ๋ถ„ํ• ๋œ ์‹ฌ์ „๋„ ์‹ ํ˜ธ๋Š” 2์ฐจ์› ์‹œ๊ฐ„-์ฃผํŒŒ์ˆ˜ ์˜์—ญ์œผ๋กœ ๋ณ€ํ™˜ํ•˜๊ณ , ์‹ ํ˜ธ ๋ถ„์„์„ ์œ„ํ•ด ์ „์ดํ•™์Šต ๋ชจ๋ธ์„ GoogleNet, ResNet-101์„ ์‚ฌ์šฉํ•˜์—ฌ ์„ฑ๋Šฅ ๊ฒฐ๊ณผ๋ฅผ ํ™•์ธํ•œ๋‹ค.

๋ณธ ๋…ผ๋ฌธ์€ 2์žฅ์€ ๊ด€๋ จ ์—ฐ๊ตฌ๋กœ ๋ฒค์น˜๋งˆํ‚น ๋ฐ์ดํ„ฐ๋ฒ ์ด์Šค, ์‹ฌ์ „๋„์˜ ํŠน์ง•์„ ์ถ”์ถœํ•˜๋Š” ๋ฐฉ๋ฒ•๊ณผ ์‹ฌ์ „๋„๋ฅผ ์ด์šฉํ•œ ๋”ฅ๋Ÿฌ๋‹๊ธฐ๋ฐ˜ ๊ฐœ์ธ์‹๋ณ„์— ๋Œ€ํ•ด ๊ธฐ์ˆ ํ•˜๊ณ , 3์žฅ์€ ์‹œ๊ฐ„-์ฃผํŒŒ์ˆ˜ ๋ณ€ํ™˜ ๋ฐฉ๋ฒ•์— ๋Œ€ํ•ด ๊ธฐ์ˆ ํ•œ๋‹ค. 4์žฅ์€ ์‹œ๊ฐ„-์ฃผํŒŒ์ˆ˜ ๋ณ€ํ™˜๊ธฐ๋ฐ˜ 2์ฐจ์› ํ•ฉ์„ฑ๊ณฑ ์‹ ๊ฒฝ๋ง์— ๋Œ€ํ•ด ๊ธฐ์ˆ ํ•œ๋‹ค. 5์žฅ์€ ์‹คํ—˜ ๊ฒฐ๊ณผ, 6์žฅ์€ ๊ฒฐ๋ก ์— ๋Œ€ํ•ด ๊ธฐ์ˆ ํ•œ๋‹ค.

2. ๊ด€๋ จ ์—ฐ๊ตฌ

2.1 ์‹ฌ์ „๋„ ๋ฐ์ดํ„ฐ๋ฒ ์ด์Šค

์‹ฌ์ „๋„ ๋ฐ์ดํ„ฐ๋ฒ ์ด์Šค๋Š” ๊ฐœ์ธ์ธ์‹๊ณผ ์งˆ๋ณ‘๋ถ„๋ฅ˜ ๋“ฑ์„ ์œ„ํ•ด ์‚ฌ์šฉ๋˜๊ณ  ์žˆ์œผ๋ฉฐ, ์ฃผ๋กœ ์‚ฌ์šฉ๋˜๊ณ  ์žˆ๋Š” ๊ณต๊ฐœ ๋ฐ์ดํ„ฐ๋ฒ ์ด์Šค์— ๋Œ€ํ•ด ์„ค๋ช…ํ•œ๋‹ค. MIT-BIH Arrhythmia Database[11]๋Š” BIH ๋ถ€์ •๋งฅ ์—ฐ๊ตฌ์†Œ์—์„œ ์—ฐ๊ตฌํ•œ 32์„ธ์—์„œ 89์„ธ ์‚ฌ์ด์˜ ๋‚จ์„ฑ 25๋ช…๊ณผ ์—ฌ์„ฑ 22๋ช…์œผ๋กœ 47๋ช…์˜ ํ”ผํ—˜์ž๊ฐ€ ํฌํ•จ๋˜์–ด ์žˆ๋‹ค. ๋˜ํ•œ 2์ฑ„๋„์˜ ์‹ ํ˜ธ๋กœ 30๋ถ„ ๊ธธ์ด์˜ 48๊ฐœ๋กœ ๊ตฌ์„ฑ๋˜์–ด ์žˆ๋‹ค. ๋ฐ์ดํ„ฐ๋ฒ ์ด์Šค๋Š” ๋ณด์Šคํ„ด์˜ Beth Israel Hospital ๋ณ‘์›์—์„œ ์ž…์› ํ™˜์ž์™€ ์™ธ๋ž˜ ํ™˜์ž์˜ ์‹ฌ์ „๋„ ์‹ ํ˜ธ๋กœ๋ถ€ํ„ฐ ์ˆ˜์ง‘๋˜์—ˆ๋‹ค. PTB Database[10]๋Š” ๋…์ผ์˜ ๊ตญ๋ฆฝ ๊ณ„์ธก ์—ฐ๊ตฌ์†Œ์—์„œ ์ˆ˜์ง‘๋œ ๋ฐ์ดํ„ฐ๋กœ 17์„ธ์—์„œ 87์„ธ ์‚ฌ์ด์˜ ๋‚จ์„ฑ 209๋ช…, ์—ฌ์„ฑ 81๋ช…์œผ๋กœ 290๋ช…์˜ ํ”ผํ—˜์ž๊ฐ€ ํฌํ•จ๋˜์–ด ์žˆ๋‹ค. 15๊ฐœ์˜ ๋ฆฌ๋“œ๋กœ ์ธก์ •๋˜์—ˆ์œผ๋ฉฐ, ๋ฐ์ดํ„ฐ๋ฒ ์ด์Šค๋Š” ์‹ฌ์žฅ์งˆํ™˜์„ ๊ฐ€์ง€๊ณ  ์žˆ๋Š” ํ”ผํ—˜์ž์˜ ์‹ ํ˜ธ์™€ ๊ฑด๊ฐ•ํ•œ ํ”ผํ—˜์ž์˜ ์‹ ํ˜ธ๋ฅผ ํฌํ•จํ•˜๊ณ  ์žˆ๋‹ค. ECG-ID[12]๋Š” 13์„ธ์—์„œ 75์„ธ ์‚ฌ์ด์˜ ๋‚จ์„ฑ 44๋ช…, ์—ฌ์„ฑ 46๋ช…์œผ๋กœ 90๋ช…์˜ ํ”ผํ—˜์ž๋กœ ํฌํ•จ๋˜์–ด ์žˆ๋‹ค. ์‹ ํ˜ธ๋Š” ๋ฆฌ๋“œ I์—์„œ ์ธก์ •๋˜์—ˆ์œผ๋ฉฐ, ํ•œ ์‚ฌ๋žŒ๋‹น 6๊ฐœ์›” ๋™์•ˆ 2๊ฐœ์—์„œ 20๊ฐœ์˜ ์‹ ํ˜ธ๋กœ ์ด 310๊ฐœ๋ฅผ ํฌํ•จํ•œ๋‹ค.

2.2 ํŠน์ง• ์ถ”์ถœ ๋ฐฉ๋ฒ•

์‹ฌ์ „๋„ ์‹ ํ˜ธ์˜ ํŠน์ง•์„ ์ถ”์ถœํ•˜๊ธฐ ์œ„ํ•ด์„œ Handcrafted์™€ Non- Handcrafted์˜ ํŠน์ง• ๋ฐฉ๋ฒ•์„ ์‚ฌ์šฉํ•œ๋‹ค.

Handcrafted ํŠน์ง• ์ถ”์ถœ ๋ฐฉ๋ฒ•์€ ์‚ฌ๋žŒ์ด ์ง์ ‘ ์‹ฌ์ „๋„ ์‹ ํ˜ธ์˜ ์‹œ๊ฐ„ ๊ฐ„๊ฒฉ, ์ง„ํญ, ๊ฐ๋„, ํ”ผํฌ(P, Q, R, S, T) ๋“ฑ์—์„œ ์‹ฌ์ง„๋„์˜ ํŠน์„ฑ์ ๊ธฐ๋ฐ˜์œผ๋กœ ํŠน์ง•์„ ์ถ”์ถœํ•œ๋‹ค[8][9]. Israel[13]์€ ์‹ฌ์ „๋„๊ธฐ๋ฐ˜ ๊ฐœ์ธ์‹๋ณ„์„ ํ•˜๊ธฐ ์œ„ํ•ด P, R, S ํ”ผํฌ์™€ ์‹œ๊ฐ„ ๊ฐ„๊ฒฉ์„ ํŠน์ง•์œผ๋กœ 15๊ฐœ์˜ ํŠน์ง•์„ ์ถ”์ถœํ–ˆ๋‹ค. Jahiruzzaman[14]๋Š” MIT-BIH Arrhythmia Database๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ์‹ฌ์ „๋„ ์‹ ํ˜ธ์— ์‹œ๊ฐ„ ์˜์—ญ์˜ ํŠน์ง• ์ถ”์ถœ์„ ํ•˜๊ธฐ ์œ„ํ•ด Continuous Wavelet Transform(CWT)๋ฅผ ์‚ฌ์šฉํ–ˆ๋‹ค. ๋˜ํ•œ ์•”ํ˜ธํ™” ๊ธฐ์ˆ ์„ ์ ์šฉํ•œ ํ›„ ID Matching์„ ์‚ฌ์šฉํ•˜์—ฌ ๊ฐœ์ธ์‹๋ณ„์„ ํ•œ๋‹ค.

Non-Handcrafted ํŠน์ง• ๋ฐฉ๋ฒ•์€ Handcrafted๊ณผ ๋ฐ˜๋Œ€๋กœ ํ•™์Šต์„ ํ†ตํ•ด ํŠน์ง•์„ ์ถ”์ถœํ•˜๋Š” ๋ฐฉ๋ฒ•์ด๋‹ค. ๋˜ํ•œ ์‹ฌ์ „๋„ ์‹ ํ˜ธ์˜ ํ˜•ํƒœ๋ฅผ ๊ธฐ๋ฐ˜์œผ๋กœ ํŠน์„ฑ์ ์„ ์‚ฌ์šฉํ•˜์ง€ ์•Š๊ธฐ ๋•Œ๋ฌธ์— ์‹ฌ์ „๋„ ์‹ ํ˜ธ์˜ ํŠน์ง• ์ถ”์ถœ์„ ์œ„ํ•ด R ํ”ผํฌ๋ฅผ ๊ฒ€์ถœํ•˜์—ฌ ์ผ์ •ํ•œ ๊ฐ„๊ฒฉ์œผ๋กœ ๋ถ„ํ• ํ•˜๋Š” ๊ณผ์ •์ด ํ•„์š”ํ•˜๋‹ค. Abdeldayem[15]์€ ์‹ฌ์ „๋„๋ฅผ ๊ธฐ๋ฐ˜์œผ๋กœ ๊ฐœ์ธ ์‹๋ณ„ํ•˜๊ธฐ ์œ„ํ•ด ๋‹ค์„ฏ ๊ฐ€์ง€์˜ ์ƒˆ๋กœ์šด ์ ‘๊ทผ ๋ฐฉ์‹์„ ์ œ์•ˆํ–ˆ๋‹ค. ์ œ์•ˆ๋œ ์ฒซ ๋ฒˆ์งธ๋Š” ์‹ฌ์ „๋„ ์‹ ํ˜ธ ๊ตฌ๋ณ„์„ ์œ„ํ•ด ์ฃผ๊ธฐ์ ์ธ ํŠน์ง•์„ ์‚ฌ์šฉํ•˜๋Š” ๋ฐฉ๋ฒ•์ด๋‹ค. ๋‘ ๋ฒˆ์งธ๋Š” ๋ณต์žกํ•œ ๊ณ„์‚ฐ ๊ฐ์†Œ์™€ ์„ฑ๋Šฅ ํ–ฅ์ƒ์„ ์œ„ํ•ด ์‹ฌ์ „๋„ ์‹ ํ˜ธ์— ์ผ์ •ํ•œ ๊ธฐ๊ฐ„์œผ๋กœ ๋ถ„ํ• ํ•˜๋Š” ๋ฐฉ๋ฒ•์„ ์‚ฌ์šฉํ•œ๋‹ค. ๋‘ ๋ฒˆ์งธ ๋ฐฉ๋ฒ•์€ ์ฒซ ๋ฒˆ์งธ์˜ ๋ฐฉ๋ฒ•์—์„œ์˜ ์‹ฌ์ „๋„์˜ ์ฃผ๊ธฐ์ ์ธ ํŠน์ง•์ด ํฌํ•จํ•ด์•ผ ํ•œ๋‹ค. ์„ธ ๋ฒˆ์งธ๋Š” ๊ธฐ์กด ๋…ผ๋ฌธ๊ณผ ๋Œ€์กฐ์ ์œผ๋กœ ์žก์Œ ์ œ๊ฑฐ ๋‹จ๊ณ„๋ฅผ ์‚ฌ์šฉํ•˜์ง€ ์•Š๋Š” ๊ฒƒ์ด๋‹ค. ๋„ค ๋ฒˆ์งธ๋Š” ์‹ ํ˜ธ๋ฅผ ์ฃผํŒŒ์ˆ˜ ์˜์—ญ์œผ๋กœ ๋ณ€ํ™˜ํ•˜์—ฌ ๋†’์€ ์ •ํ™•๋„๋ฅผ ๋‚˜ํƒ€๋‚ด๋Š” 2D-CNN์„ ์‚ฌ์šฉํ•œ๋‹ค. ๋‹ค์„ฏ ๋ฒˆ์งธ๋Š” 8๊ฐœ์˜ ๋ฐ์ดํ„ฐ๋ฒ ์ด์Šค๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ํ•˜๋‚˜์˜ ๋ฐ์ดํ„ฐ๋กœ ๊ฒฐํ•ฉํ•˜๊ณ , 488๋ช…์˜ ํ”ผํ—˜์ž์˜ ์‹ฌ์ „๋„ ์‹ ํ˜ธ๋ฅผ ์‚ฌ์šฉํ–ˆ๋‹ค. Labati[16]๋Š” PTB Database๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ CNN๊ธฐ๋ฐ˜ Deep- ECG๋ฅผ ์ œ์•ˆํ–ˆ๋‹ค. CNN์€ 6๊ฐœ์˜ ์ปจ๋ณผ๋ฃจ์…˜ ๊ณ„์ธต, 1๊ฐœ์˜ ๋“œ๋กญ์•„์›ƒ ๊ณ„์ธต, ์™„์ „ ์—ฐ๊ฒฐ ๊ณ„์ธต, ์†Œํ”„ํŠธ๋งฅ์Šค ๊ณ„์ธต์œผ๋กœ ๊ตฌ์„ฑํ•˜์—ฌ ๊ฐœ์ธ์‹๋ณ„์„ ์ง„ํ–‰ํ–ˆ๋‹ค. Y. H. Byeon[17]์€ 5๊ฐœ์˜ ์‹œ๊ฐ„-์ฃผํŒŒ์ˆ˜ ํ‘œํ˜„์œผ๋กœ CNN์„ ์‚ฌ์šฉํ•˜์—ฌ ์‹ฌ์ „๋„๊ธฐ๋ฐ˜ ์ƒ์ฒด ์ธ์‹ ์„ฑ๋Šฅ์„ ํ™•์ธํ•˜์˜€๋‹ค. ์‹œ๊ฐ„-์ฃผํŒŒ์ˆ˜ ํ‘œํ˜„ ๋ฐฉ๋ฒ•์œผ๋กœ scalogram, spectrogram, mel spectrogram, log spectrogram, spectrogram, MFCC๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ์ „์ดํ•™์Šต ๋ชจ๋ธ์ธ VGGNet, ResNet, Xception, DenseNet์„ ์‚ฌ์šฉํ–ˆ๋‹ค.

2.3 ๋”ฅ๋Ÿฌ๋‹๊ธฐ๋ฐ˜ ์‹ฌ์ „๋„๋ฅผ ์ด์šฉํ•œ ๊ฐœ์ธ์‹๋ณ„

์ตœ๊ทผ ์‹ฌ์ „๋„๋ฅผ ๊ธฐ๋ฐ˜์œผ๋กœ ๋”ฅ๋Ÿฌ๋‹์„ ์‚ฌ์šฉํ•˜์—ฌ ๊ฐœ์ธ์‹๋ณ„ ๊ด€๋ จ ์—ฐ๊ตฌ๊ฐ€ ํ™œ๋ฐœํžˆ ์ง„ํ–‰๋˜๊ณ  ์žˆ๋‹ค. ์‹ฌ์ „๋„๋ฅผ ์ด์šฉํ•œ ๊ฐœ์ธ ์‚ฌ์šฉ์ž ์ธ์‹ ์—ฐ๊ตฌ๋Š” ์ตœ์ดˆ๋กœ Biel[18]์— ์˜ํ•ด ์—ฐ๊ตฌ๊ฐ€ ์ง„ํ–‰๋˜์—ˆ๋‹ค. D. Jyotishi[19]๋Š” ์‹ฌ์ „๋„ ์‹ ํ˜ธ๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ๊ฐœ์ธ์„ ์‹๋ณ„ํ•˜๊ธฐ ์œ„ํ•ด LSTM์˜ ์…€ ์ถœ๋ ฅ์„ ํ•ฉ์‚ฐํ•˜์—ฌ ๋ถ„๋ฅ˜ํ•˜๋Š” ๋ฐฉ๋ฒ•์„ ์ œ์•ˆํ–ˆ๋‹ค. J. S. Kim[20]๋Š” ์‹ฌ์ „๋„ ์‹ ํ˜ธ์˜ ์ฃผ๊ธฐ์ ์ธ ํŠน์ง•์„ ์ด์šฉํ•˜์—ฌ 2D ์ปคํ”Œ๋ง ์ด๋ฏธ์ง€๋ฅผ ์‚ฌ์šฉํ•œ ๊ฐœ์ธ์‹๋ณ„์„ ์ œ์•ˆํ–ˆ๋‹ค. 2D ์ปคํ”Œ๋ง ์ด๋ฏธ์ง€๋ฅผ ์œ„ํ•ด 12๊ฐœ์˜ ์ปจ๋ณผ๋ฃจ์…˜ ๊ณ„์ธต, 6๊ฐœ์˜ ์ตœ๋Œ€ ํ’€๋ง ๊ณ„์ธต์œผ๋กœ ๊ตฌ์„ฑ๋œ CNN ๊ตฌ์กฐ๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ์‹ฌ์ „๋„๋ฅผ ์ด์šฉํ•œ ์‚ฌ์šฉ์ž ์ธ์‹ ๋ฐฉ๋ฒ•์„ ์ œ์•ˆํ–ˆ๋‹ค. M. Hammad[21]๋Š” ECG๊ธฐ๋ฐ˜ ์ธ์ฆ์„ ์œ„ํ•ด ์‹ฌ์ธต ์‹ ๊ฒฝ๋ง ๋ชจ๋ธ์„ ์ œ์•ˆํ–ˆ๋‹ค. ์ฒซ ๋ฒˆ์งธ๋Š” 4๊ฐœ์˜ ํ•ฉ์„ฑ๊ณฑ ๊ณ„์ธต๊ณผ 2๊ฐœ์˜ ์ตœ๋Œ€ ํ’€๋ง ๊ณ„์ธต, 2๊ฐœ์˜ ์™„์ „ ์—ฐ๊ฒฐ ๊ณ„์ธต, 1๊ฐœ์˜ ์ตœ๋Œ€ ํ’€๋ง ๊ณ„์ธต์œผ๋กœ ๊ตฌ์„ฑ๋œ 1D-CNN ๋ชจ๋ธ์„ ์„ค๊ณ„ํ•˜์˜€๋‹ค. CNN์˜ ํ•ฉ์„ฑ๊ณฑ ์—ฐ์‚ฐ์€ ์ด๋ฏธ์ง€ ๋ฐ์ดํ„ฐ์˜ ํ˜•ํƒœํ•™์ ์ธ ํŠน์ง•์„ ํšจ๊ณผ์ ์œผ๋กœ ์ถ”์ถœํ•  ์ˆ˜ ์žˆ๋‹ค. ๋‘ ๋ฒˆ์งธ ๋ชจ๋ธ์€ 2๊ฐœ์˜ ํ•ฉ์„ฑ๊ณฑ ๊ณ„์ธต, ์ •๊ทœํ™” ๊ณ„์ธต, ReLU ๊ณ„์ธต, ๋“œ๋กญ์•„์›ƒ ๊ณ„์ธต์œผ๋กœ ๊ตฌ์„ฑ๋œ ์ฒซ ๋ฒˆ์งธ ๊ณ„์ธต๊ณผ 2๊ฐœ์˜ ์ •๊ทœํ™” ๊ณ„์ธต, 2๊ฐœ์˜ ReLU ๊ณ„์ธต, 2๊ฐœ์˜ ๋“œ๋กญ์•„์›ƒ ๊ณ„์ธต, 2๊ฐœ์˜ ํ•ฉ์„ฑ๊ณฑ ๊ณ„์ธต์œผ๋กœ ๊ตฌ์„ฑ๋œ ๋‘ ๋ฒˆ์งธ ๊ณ„์ธต์˜ ์ถœ๋ ฅ์„ ํ•ฉ๋ณ‘ํ•˜์—ฌ Attention์˜ ์ž…๋ ฅ์œผ๋กœ ์‚ฌ์šฉ๋˜๋Š” ResNet-Attention ๋ชจ๋ธ์„ ์„ค๊ณ„ํ–ˆ๋‹ค. Attention ๊ตฌ์กฐ๋Š” 2๊ฐœ์˜ Dense ๊ณ„์ธต, ReLU ๊ณ„์ธต, ์†Œํ”„ํŠธ๋งฅ์Šค ๊ณ„์ธต์„ ํ†ตํ•ด ์‚ฌ์šฉ์ž ์ธ์ฆ ์„ฑ๋Šฅ์„ ํ™•์ธํ•œ๋‹ค.

3.์‹œ๊ฐ„-์ฃผํŒŒ์ˆ˜ ๋ณ€ํ™˜ ๋ฐฉ๋ฒ•

์‹ฌ์ „๋„, ๊ทผ์ „๋„, ๋‡Œ์ „๋„์™€ ๊ฐ™์€ ์ƒ๋ฆฌํ•™์  ์‹ ํ˜ธ๋Š” ์žก์Œ์˜ ์˜ํ–ฅ์„ ๋งŽ์ด ๋ฐ›๊ธฐ ๋•Œ๋ฌธ์— ์‹ ํ˜ธ๋ฅผ ๋ถ„์„ํ•˜๊ธฐ ์œ„ํ•ด 1์ฐจ์› ์‹ ํ˜ธ๋ฅผ ์‹œ๊ฐ„-์ฃผํŒŒ์ˆ˜ ์˜์—ญ์œผ๋กœ ๋ณ€ํ™˜ํ•˜์—ฌ 2์ฐจ์› ์ด๋ฏธ์ง€๋กœ ๋‚˜ํƒ€๋‚ธ๋‹ค[22]. ์‹œ๊ฐ„-์ฃผํŒŒ์ˆ˜ ์˜์—ญ์˜ ๋ณ€ํ™˜์„ ํ†ตํ•ด ์‹ฌ์ „๋„ ์‹ ํ˜ธ๋Š” Short-Time Fourier Transform(STFT), Fourier Synchrosqueezed Transform(FSST), Wavelet Synchrosqueezed Transform(WSST)๋กœ ํ‘œํ˜„๋œ ์ด๋ฏธ์ง€๋ฅผ ์‚ฌ์šฉํ•œ๋‹ค.

3.1 STFT

STFT๋Š” ์‹œ๊ฐ„์ด ์ง€๋‚จ์— ๋”ฐ๋ผ ๋ณ€ํ•˜๋Š” ์‹ ํ˜ธ๋ฅผ ์ผ์ •ํ•œ ๊ธธ์ด๋กœ ๋ถ„ํ• ํ•˜์—ฌ ํ‘ธ๋ฆฌ์— ๋ณ€ํ™˜์— ์ ์šฉ์‹œํ‚จ๋‹ค. ํ‘ธ๋ฆฌ์— ๋ณ€ํ™˜์€ ์‹œ๊ฐ„ ์˜์—ญ์˜ ์‹ ํ˜ธ๋ฅผ ์ฃผํŒŒ์ˆ˜๋กœ ๋ถ„ํ•ดํ•˜๋Š” ๋ณ€ํ™˜์ด๋‹ค. STFT๋Š” ์‹ ํ˜ธ๋ฅผ ๋ถ„ํ• ํ•˜๋Š” ๊ธธ์ด์— ๋”ฐ๋ผ ์‹œ๊ฐ„ ๋˜๋Š” ์ฃผํŒŒ์ˆ˜์˜ ๋ถ„ํ•ด๋Šฅ์ด ์ข‹์Œ์„ ํ™•์ธํ•  ์ˆ˜ ์žˆ๋‹ค. ์ˆ˜์‹ (1)์€ ์‹ ํ˜ธ๋ฅผ ๋ถ„ํ• ํ•˜๋Š” ๊ณผ์ •์ด๊ณ , ์ˆ˜์‹ (2)๋Š” ํ‘ธ๋ฆฌ์— ๋ณ€ํ™˜์˜ ๊ณ„์‚ฐ ๊ณผ์ •์œผ๋กœ, $x(t)$๋Š” ์‹œ๊ฐ„ t์— ๋Œ€ํ•œ ์‹ ํ˜ธ, $w(t)$๋Š” ์œˆ๋„์šฐ ํ•จ์ˆ˜, $m$์€ ์œˆ๋„์šฐ ํ•จ์ˆ˜์˜ ์ค‘์ ์„ ๋‚˜ํƒ€๋‚ธ๋‹ค[23].

(1)
$x(m,\: t)= x(t)*w(t-f)$
(2)
$x(m,\: u)=\int_{-\infty}^{\infty}x(t)w(t-m)e^{-iut}dt$

๊ทธ๋ฆผ 1์€ 128์˜ ์œˆ๋„์šฐ ๊ธธ์ด์— ๋”ฐ๋ฅธ STFT๋ฅผ ๋ณด์—ฌ์ค€๋‹ค.

๊ทธ๋ฆผ 1. 128์˜ ์œˆ๋„์šฐ ๊ธธ์ด๋ฅผ ๊ฐ€์ง„ STFT

Fig. 1. STFT with a window length of 128

../../Resources/kiee/KIEEP.2022.71.1.54/fig1.png

3.2FSST

๊ธฐ๊ณ„ ์ง„๋™, ์Œ์„ฑ ๋ฐ ์ƒ๋ฆฌํ•™์  ์‹ ํ˜ธ ๋“ฑ์€ ์ฃผํŒŒ์ˆ˜ ๋˜๋Š” ์ง„ํญ ๋ณ€์กฐ์˜ ์ค‘์ฒฉ์œผ๋กœ ํ‘œํ˜„๋  ์ˆ˜ ์žˆ๋‹ค. ์‹œ๊ฐ„-์ฃผํŒŒ์ˆ˜ ๋ถ„์„์„ ์œ„ํ•ด ๋ถ„์„ ์‹ ํ˜ธ์˜ ํ•ฉ์€ ์ˆ˜์‹ (3)๊ณผ ๊ฐ™์ด ํ‘œํ˜„ํ•  ์ˆ˜ ์žˆ๋‹ค. $f(t)$์™€ $ฯ•_{a}(t)$๋Š” $X_{a}(t)$์˜ ๋ถ„์„ ์‹ ํ˜ธ์˜ ์‹œ๊ฐ„-์ฃผํŒŒ์ˆ˜ ์ง„ํญ๊ณผ ์œ„์ƒ, $j$๋Š” $\sqrt{-1}$, $N$์€ ๋ถ„์„ ์‹ ํ˜ธ ์ˆ˜์ด๋‹ค. FSST๋Š” STFT์˜ ๊ธฐ๋ฐ˜์œผ๋กœ ์„ ๋ช…ํ•œ ์‹œ๊ฐ„-์ฃผํŒŒ์ˆ˜๋ฅผ ์ƒ์„ฑํ•œ๋‹ค[24]. ์ˆ˜์‹ (4)์™€ ๊ฐ™์ด ์ŠคํŽ™ํŠธ๋Ÿผ ์œˆ๋„์šฐ g์™€ ํ•จ์ˆ˜ $f$๋ฅผ ๊ณ„์‚ฐํ•˜์—ฌ STFT์„ ๋ณ€ํ™˜ํ•œ๋‹ค. FSST๋Š” STFT๋ฅผ ๊ตฌํ•˜๋Š” ์ˆ˜์‹์—์„œ $e^{j2\pi\eta t}$๋ฅผ ์ถ”๊ฐ€ํ•˜์—ฌ STFT๋กœ ๋ณ€ํ™˜๋œ ๊ฐ’์ด ์••์ถ•๋˜์–ด ํ‘œํ˜„๋œ๋‹ค. ์ˆ˜์‹ (4)๋ถ€ํ„ฐ ์ˆ˜์‹ (6)์€ FSST์˜ ๊ณ„์‚ฐ ๊ณผ์ •์„ ๋‚˜ํƒ€๋‚ธ๋‹ค. ๊ทธ๋ฆผ 2๋Š” 1์ฐจ์›์ธ ์‹ฌ์ „๋„ ์‹ ํ˜ธ๋ฅผ ์‹œ๊ฐ„-์ฃผํŒŒ์ˆ˜ ํ‘œํ˜„์ธ FSST๋กœ ๋‚˜ํƒ€๋‚ธ ๊ฒƒ์„ ๋ณด์—ฌ์ค€๋‹ค.

(3)
$f(t)=\sum_{a=1}^{A}f_{a}(t)=\sum_{a=1}^{A}X_{a}(t)e^{j2\pi ฯ•_{a}(t)}$
(4)
$K_{g}f(t,\: \eta)=\int_{-\infty}^{\infty}f(s)g(s-t)e^{-j2\pi\eta(s-t)}ds$
(5)
$V_{g}f(t,\: \omega)=\int_{-\infty}^{\infty}H_{g}f(t,\: \eta)\delta(\omega -\omega_{g}f(t,\: \eta))d\eta$
(6)
$\omega_{g}f(x,\: \eta)=\eta -\dfrac{1}{j2\pi}\dfrac{H_{\partial_{g}/\partial_{x}}f(x,\: \eta)}{H_{g}f(x,\: \eta)}$

๊ทธ๋ฆผ 2. FSST๋กœ ํ‘œํ˜„๋œ ์‹ฌ์ „๋„ ์‹ ํ˜ธ

Fig. 2. An electrocardiogram signal expressed as FSST

../../Resources/kiee/KIEEP.2022.71.1.54/fig2.png

3.3WSST

WSST๋Š” ์Œ์„ฑ, ์ƒ๋ฆฌํ•™์  ์‹ ํ˜ธ, ๊ธฐ๊ณ„ ๋“ฑ๊ณผ ๊ฐ™์€ ์ง„๋™์ด ์žˆ๋Š” ๋‹ค์„ฑ๋ถ„ ์‹ ํ˜ธ๋ฅผ ๋ถ„์„ํ•˜๋Š”๋ฐ ์œ ์šฉํ•œ ์‹œ๊ฐ„-์ฃผํŒŒ์ˆ˜ ํ‘œํ˜„ ๋ฐฉ๋ฒ•์ด๋‹ค. ๋˜ํ•œ WSST๋Š” ์‹ ํ˜ธ ์—๋„ˆ์ง€๋ฅผ ์ฃผํŒŒ์ˆ˜๋กœ ์žฌํ• ๋‹นํ•˜๋Š” ํ‘œํ˜„ ๋ฐฉ๋ฒ•์œผ๋กœ mather ์›จ์ด๋ธ”๋ฆฟ์œผ๋กœ ์‚ฐ๋ž€๋œ ํšจ๊ณผ๋ฅผ ๋ณด์ƒํ•  ์ˆ˜ ์žˆ๋‹ค[25]. WSST์˜ ์•Œ๊ณ ๋ฆฌ์ฆ˜ ์ˆœ์„œ๋Š” ๋‹ค์Œ๊ณผ ๊ฐ™์ด ๋‚˜ํƒ€๋‚ธ๋‹ค.

1. ์ž…๋ ฅ ์‹ ํ˜ธ์˜ CWT์„ ๊ตฌํ•˜๊ธฐ ์œ„ํ•ด ์ˆ˜์‹ (7)๊ณผ ๊ฐ™์ด ๊ณ„์‚ฐํ•œ๋‹ค. C๋Š” CWT๋ฅผ ์˜๋ฏธํ•˜๊ณ , n๊ณผ b๋Š” ์Šค์ผ€์ผ๋ง๊ณผ ์‹œํ”„ํŠธ ํŒŒ๋ผ๋ฏธํ„ฐ๋ฅผ ์˜๋ฏธํ•œ๋‹ค.

(7)
$C(n,\: b)=\dfrac{1}{\sqrt{n}}\int_{-\infty}^{\infty}h(t)\psi(\dfrac{t-b}{n})dt$

2. Synchrosqueezing๋กœ ํ‘œํ˜„ํ•˜๊ธฐ ์œ„ํ•ด ์ˆ˜์‹ (8)์„ ๊ณ„์‚ฐํ•˜๊ณ , CWT์˜ ์ถœ๋ ฅ์—์„œ ์ˆœ๊ฐ„ ์ฃผํŒŒ์ˆ˜ ์ •๋ณด๋ฅผ ์ถ”์ถœํ•œ๋‹ค.

(8)
$h_{c}= -i\dfrac{\dfrac{d C(n,\: b)}{db}}{C(n,\: b)}$

3. CWT์˜ ๊ฐ’์„ ์••์ถ•ํ•˜๊ณ , ์ˆœ๊ฐ„ ์ฃผํŒŒ์ˆ˜ ์ •๋ณด ๊ฐ’์€ ์‹œ๊ฐ„-์ฃผํŒŒ์ˆ˜ ์˜์—ญ์˜ ์ค‘์‹ฌ์—์„œ ๋‹จ์ผ ๊ฐ’์œผ๋กœ ์žฌํ• ๋‹น๋œ๋‹ค. ์žฌํ• ๋‹น์— ์˜ํ•ด WSST๋Š” ์„ ๋ช…ํ•œ ์ถœ๋ ฅ์„ ์–ป๋Š”๋‹ค.

๊ทธ๋ฆผ 3. WSST๋กœ ํ‘œํ˜„๋œ ์‹ฌ์ „๋„ ์‹ ํ˜ธ

Fig. 3. An electrocardiogram signal expressed as WSST

../../Resources/kiee/KIEEP.2022.71.1.54/fig3.png

4. ์‹œ๊ฐ„-์ฃผํŒŒ์ˆ˜ ๋ณ€ํ™˜๊ธฐ๋ฐ˜ 2์ฐจ์› ํ•ฉ์„ฑ๊ณฑ ์‹ ๊ฒฝ๋ง

4.1ํ•ฉ์„ฑ๊ณฑ ์‹ ๊ฒฝ๋ง

๋”ฅ๋Ÿฌ๋‹์€ ๊ธฐ๊ณ„ํ•™์Šต์˜ ํŠน์ •ํ•œ ํ•œ ๋ถ„์•ผ๋กœ, ์—ฌ๋Ÿฌ ์ธต์„ ๊ฐ€์ง„ ์‹ฌ์ธต ์‹ ๊ฒฝ๋ง์œผ๋กœ ๊ตฌ์„ฑ๋œ๋‹ค. ์‹ฌ์ธต ์‹ ๊ฒฝ๋ง์€ ์ž…๋ ฅ ๊ณ„์ธต, 2๊ฐœ ์ด์ƒ์˜ ์€๋‹‰์ธต, ์ถœ๋ ฅ์ธต์œผ๋กœ ๊ตฌ์„ฑ๋˜๋ฉฐ, ๋†’์€ ์„ฑ๋Šฅ์„ ๊ธฐ๋Œ€ํ•  ์ˆ˜ ์žˆ๋‹ค.

ํ•ฉ์„ฑ๊ณฑ ์‹ ๊ฒฝ๋ง์€ ๋”ฅ๋Ÿฌ๋‹์˜ ์•„ํ‚คํ…์ฒ˜๋กœ ์ฃผ๋กœ ์ด๋ฏธ์ง€ ๋ถ„์•ผ์—์„œ ๋งŽ์ด ์‚ฌ์šฉ๋˜๊ณ  ์žˆ๋‹ค. ๋˜ํ•œ ํ•ฉ์„ฑ๊ณฑ ์‹ ๊ฒฝ๋ง์€ ๊ทธ๋ฆผ 4์™€ ๊ฐ™์ด ํ•ฉ์„ฑ๊ณฑ ๊ณ„์ธต, ReLU ๊ณ„์ธต, ํ’€๋ง ๊ณ„์ธต์ด ๋ฐ˜๋ณต๋˜๋ฉฐ, ์ž…๋ ฅ๋œ ๋ฐ์ดํ„ฐ๋Š” ํ•ฉ์„ฑ๊ณฑ ์—ฐ์‚ฐ์„ ํ†ตํ•ด ํŠน์ง•์ด ์ถ”์ถœ๋œ๋‹ค.

๊ทธ๋ฆผ 4. ํ•ฉ์„ฑ๊ณฑ ์‹ ๊ฒฝ๋ง์˜ ๊ตฌ์กฐ

Fig. 4. The structure of the convolutional neural network

../../Resources/kiee/KIEEP.2022.71.1.54/fig4.png

4.2์‹œ๊ฐ„-์ฃผํŒŒ์ˆ˜ ๋ณ€ํ™˜๊ธฐ๋ฐ˜ ํ•ฉ์„ฑ๊ณฑ ์‹ ๊ฒฝ๋ง

1์ฐจ์› ์‹ฌ์ „๋„ ์‹ ํ˜ธ๋ฅผ 2์ฐจ์› ์‹œ๊ฐ„-์ฃผํŒŒ์ˆ˜ ๋ณ€ํ™˜์„ ํ†ตํ•ด STFT, FSST, WSST์˜ ์ด๋ฏธ์ง€๋กœ ํ‘œํ˜„ํ•˜์—ฌ ์ด๋ฏธ์ง€ ๋ถ„๋ฅ˜ ์„ฑ๋Šฅ์ด ๋†’์€ ํ•ฉ์„ฑ๊ณฑ ์‹ ๊ฒฝ๋ง์„ ์‚ฌ์šฉํ•œ๋‹ค. ํ•ฉ์„ฑ๊ณฑ ์‹ ๊ฒฝ๋ง์€ ์ง์ ‘ ๋ชจ๋ธ์„ ์„ค๊ณ„ํ•˜์—ฌ ๋ชจ๋ธ์„ ํ›ˆ๋ จ์‹œํ‚ค๊ธฐ ์œ„ํ•ด ๋งŽ์€ ๋ฐ์ดํ„ฐ ์ˆ˜, ๋งŽ์€ ์‹œ๊ฐ„, ๋ณต์žกํ•œ ๋‹จ์ ์ด ์žˆ๊ธฐ ๋•Œ๋ฌธ์— ํ•ฉ์„ฑ๊ณฑ ์‹ ๊ฒฝ๋ง์˜ ๋ชจ๋ธ๋กœ ์ „์ดํ•™์Šต์„ ์‚ฌ์šฉํ•œ๋‹ค. ์ „์ดํ•™์Šต์€ ๊ธฐ์กด์— ๋งŒ๋“ค์–ด์ง„ ๋ชจ๋ธ์„ ์‚ฌ์šฉํ•˜์—ฌ ์ƒˆ๋กœ์šด ๋ฐ์ดํ„ฐ๋ฅผ ํ•™์Šต ์‹œํ‚ค๋Š” ๋ฐฉ๋ฒ•์ด๋‹ค. ๋˜ํ•œ ์ „์ดํ•™์Šต์€ ์ง์ ‘ ์„ค๊ณ„ํ•œ ํ•ฉ์„ฑ๊ณฑ ์‹ ๊ฒฝ๋ง๊ณผ ๋ฐ˜๋Œ€๋กœ ์ž‘์€ ๋ฐ์ดํ„ฐ ์ˆ˜์—๋„ ํ•™์Šต์ด ๊ฐ€๋Šฅํ•˜๋‹ค๋Š” ์žฅ์ ์ด ์žˆ๋‹ค. 2์ฐจ์› ๋ณ€ํ™˜๊ธฐ๋ฐ˜ ํ•ฉ์„ฑ๊ณฑ ์‹ ๊ฒฝ๋ง์€ GoogleNet, ResNet-101๋ฅผ ์‚ฌ์šฉํ•œ๋‹ค. GoogleNet์€ ํ•™์Šต์ด ๊ฐ€๋Šฅํ•œ 22๊ฐœ์˜ ๊ณ„์ธต๊ณผ 9๊ฐœ์˜ ์ธ์…‰์…˜ ๋ชจ๋“ˆ์œผ๋กœ ๊ตฌ์„ฑ๋œ ์‹ฌ์ธต ๋„คํŠธ์›Œํฌ์ด๋‹ค. ๋˜ํ•œ ์ธ์…‰์…˜ ๋ชจ๋“ˆ์„ ํฌํ•จํ•œ ๋ณ‘๋ ฌ ํ•ฉ์„ฑ๊ณฑ ํ•„ํ„ฐ๋กœ ์ถœ๋ ฅ์ด ์—ฐ๊ฒฐ๋˜์–ด ๊ตฌ์„ฑ๋˜์–ด ์žˆ๋‹ค[26]. ๊ทธ๋ฆผ 6์€ ์ธ์…‰์…˜ ๋ชจ๋“ˆ๋กœ 1*1 ํ•ฉ์„ฑ๊ณฑ, 1*1 ํ•ฉ์„ฑ๊ณฑ + 3*3 ํ•ฉ์„ฑ๊ณฑ, 1*1 ํ•ฉ์„ฑ๊ณฑ + 5*5 ํ•ฉ์„ฑ๊ณฑ, 3*3 ์ตœ๋Œ€ ํ’€๋ง + 1*1 ํ•ฉ์„ฑ๊ณฑ์œผ๋กœ 4๊ฐ€์ง€ ์—ฐ์‚ฐ์ด ์ˆ˜ํ–‰๋œ๋‹ค. ์ด๋Ÿฌํ•œ ์—ฐ์‚ฐ๋“ค์€ ํŒŒ๋ผ๋ฏธํ„ฐ ์ˆ˜๋ฅผ ๊ฐ์†Œ์‹œํ‚ค๊ณ , ์ฑ„๋„์˜ ์ˆ˜๋ฅผ ์กฐ์ ˆํ•จ์œผ๋กœ์„œ ํ•„์š”ํ•œ ์—ฐ์‚ฐ๋Ÿ‰์„ ์ค„์ด๊ธฐ ์œ„ํ•ด ์‚ฌ์šฉ๋œ๋‹ค. GoogleNet์€ ๊ทธ๋ฆผ 5์™€ ๊ฐ™์ด ๊ตฌ์„ฑ๋˜์–ด ์žˆ๋‹ค. ํ•˜์ง€๋งŒ GoogleNet์€ ์‹ฌ์ธต ์‹ ๊ฒฝ๋ง์œผ๋กœ, vanishing gradient๊ฐ€ ์ƒ๊ธฐ๊ธฐ ๋•Œ๋ฌธ์— ์ด๋ฅผ ๋ฐฉ์ง€ํ•˜๊ธฐ ์œ„ํ•ด ๊ณ„์‚ฐ๋œ ์†์‹ค ๊ฐ’์„ ์ถ”๊ฐ€ํ•˜๋Š” ๊ณผ์ •์œผ๋กœ ํ›ˆ๋ จ ๊ณผ์ • ์ค‘์—๋งŒ ์‚ฌ์šฉ๋˜๋Š” ๋ณด์กฐ ๋ถ„๋ฅ˜๊ธฐ๊ฐ€ ํฌํ•จ๋œ๋‹ค.

๊ทธ๋ฆผ 5. GoogleNet์˜ ๊ตฌ์กฐ

Fig. 5. The structure of GoogleNet

../../Resources/kiee/KIEEP.2022.71.1.54/fig5.png

๊ทธ๋ฆผ 6. GoogleNet์˜ ์ธ์…‰์…˜ ๋ชจ๋“ˆ

Fig. 6. GoogleNet's Inception Module

../../Resources/kiee/KIEEP.2022.71.1.54/fig6.png

ResNet-101์€ 33๊ฐœ์˜ ๊ณ„์ธต์ธ ๋ธ”๋ก์œผ๋กœ ๊ตฌ์„ฑ๋˜์–ด ์žˆ์œผ๋ฉฐ, ํ•ฉ์„ฑ๊ณฑ ๊ณ„์ธต์€ 104๊ฐœ๋ฅผ ํฌํ•จํ•˜๊ณ  ์žˆ๋‹ค. 1*1 ํ•ฉ์„ฑ๊ณฑ, 3*3 ํ•ฉ์„ฑ๊ณฑ, 1*1 ํ•ฉ์„ฑ๊ณฑ๊ณผ ๊ฐ™์€ ๋ณ‘๋ชฉ ๊ณ„์ธต์„ ํ†ตํ•ด ์—ฐ์‚ฐ์„ ์ค„์ผ ์ˆ˜ ์žˆ์œผ๋ฉฐ, Residual Connection์„ ์ถ”๊ฐ€ํ•˜์—ฌ vanishing gradient์„ ํ•ด๊ฒฐํ•  ์ˆ˜ ์žˆ๋‹ค. ์ด๋ฅผ ์ถ”๊ฐ€ํ•จ์œผ๋กœ์จ ์‹ ๊ฒฝ๋ง์˜ ๊ณ„์ธต์ด ๊นŠ์–ด ๋” ๋†’์€ ์„ฑ๋Šฅ์„ ํ™•์ธํ•  ์ˆ˜ ์žˆ๋‹ค[27]. ๊ทธ๋ฆผ 7๊ณผ ๊ทธ๋ฆผ 8์€ ๊ฐ๊ฐ ResNet-101์˜ ๊ตฌ์กฐ์™€ Residual Connection์„ ๋ณด์—ฌ์ค€๋‹ค.

๊ทธ๋ฆผ 7. ResNet-101์˜ ๊ตฌ์กฐ

Fig. 7. The structure of ResNet-101

../../Resources/kiee/KIEEP.2022.71.1.54/fig7.png

๊ทธ๋ฆผ 8. Residual Connection

Fig. 8. Residual Connection

../../Resources/kiee/KIEEP.2022.71.1.54/fig8.png

5. ์‹คํ—˜ ๊ฒฐ๊ณผ

๋ณธ ๋…ผ๋ฌธ์€ ์‹ฌ์ „๋„๊ธฐ๋ฐ˜์œผ๋กœ ๊ฐœ์ธ์‹๋ณ„์„ ํ•˜๊ธฐ ์œ„ํ•ด ๋ฒค์น˜๋งˆํ‚น ๋ฐ์ดํ„ฐ๋ฒ ์ด์Šค PTB๋ฅผ ์‚ฌ์šฉํ•˜์˜€๋‹ค. PTB ๋ฐ์ดํ„ฐ๋ฒ ์ด์Šค๋Š” 290๋ช…์˜ ํ”ผํ—˜์ž๋กœ ๊ตฌ์„ฑ๋˜์–ด ์žˆ์ง€๋งŒ ๊ทธ ์ค‘ R-peak์˜ ๊ฐ์ง€๊ฐ€ ์–ด๋ ค์šด 79๋ช…์„ ์ œ์™ธํ•˜๊ณ  211๋ช… ํ”ผํ—˜์ž์˜ ์‹ฌ์ „๋„ ์‹ ํ˜ธ๋ฅผ ์‚ฌ์šฉํ•˜์˜€๋‹ค. ํ”ผํ—˜์ž ๋‹น 120๊ฐœ์˜ ๋ฐ์ดํ„ฐ๋กœ ๊ตฌ์„ฑ๋˜๋ฉฐ, ํ•™์Šต 60๊ฐœ, ๊ฒ€์ฆ 60๊ฐœ๋ฅผ ์‚ฌ์šฉํ•˜์˜€๋‹ค.

์‹ฌ์ „๋„ ์‹ ํ˜ธ๋Š” ์žก์Œ์— ์˜ํ–ฅ์„ ํฌ๊ฒŒ ๋ฏธ์น˜๊ธฐ ๋•Œ๋ฌธ์— 1์ฐจ์› ์‹ฌ์ „๋„ ์‹ ํ˜ธ๋ฅผ 2์ฐจ์› ์ด๋ฏธ์ง€๋กœ ๋ณ€ํ™˜ํ•˜์—ฌ ํ•ฉ์„ฑ๊ณฑ ์‹ ๊ฒฝ๋ง์„ ํ†ตํ•ด ๊ฐœ์ธ์‹๋ณ„ํ•œ๋‹ค. ๊ทธ๋ฆผ 9๋Š” PTB ๋ฐ์ดํ„ฐ๋ฒ ์ด์Šค ์‹ ํ˜ธ ์ค‘ ์ž„์˜์˜ 2๋ช…์˜ ํ”ผํ—˜์ž์— ๋Œ€ํ•œ STFT, FSST, WSST๋กœ ๋ณ€ํ™˜์‹œํ‚จ ์ด๋ฏธ์ง€๋ฅผ ๋ณด์—ฌ์ค€๋‹ค. ๊ทธ๋ฆผ 9์˜ ์‹ฌ์ „๋„ ์› ์‹ ํ˜ธ์™€ ๊ฐ™์ด ์‹ฌ์ „๋„๋Š” ์‚ฌ๋žŒ๋งˆ๋‹ค ๋‹ค๋ฅด๊ฒŒ ๋‚˜ํƒ€๋‚˜๋Š” ๊ฒƒ์„ ํ™•์ธํ•  ์ˆ˜ ์žˆ๊ณ , ์‚ฌ์šฉํ•œ STFT, FSST, WSST์˜ 3๊ฐœ์˜ ๋ณ€ํ™˜ ๋ฐฉ๋ฒ•์— ๋”ฐ๋ผ ์‹ฌ์ „๋„๊ฐ€ ๋‹ค๋ฅด๊ฒŒ ๋‚˜ํƒ€๋‚ด๋Š” ๊ฒƒ์„ ํ™•์ธํ•  ์ˆ˜ ์žˆ๋‹ค.

๊ทธ๋ฆผ 9. ํ”ผํ—˜์ž์— ๋Œ€ํ•œ ์‹œ๊ฐ„-์ฃผํŒŒ์ˆ˜ ํ‘œํ˜„

Fig. 9. Time-frequency representation for a subject

../../Resources/kiee/KIEEP.2022.71.1.54/fig9.png

์‹œ๊ฐ„-์ฃผํŒŒ์ˆ˜๋กœ ๋ณ€ํ™˜๋œ 2์ฐจ์› ์ด๋ฏธ์ง€๋Š” ํ•ฉ์„ฑ๊ณฑ ์‹ ๊ฒฝ๋ง์˜ ์‹ฌ์ธต ๋ชจ๋ธ์ธ GoogleNet, ResNet-101๋ฅผ ์‚ฌ์šฉํ•˜์˜€๋‹ค. ๋ฐ์ดํ„ฐ๋Š” 224*224 ํฌ๊ธฐ๋กœ ๋ณ€๊ฒฝํ•˜์—ฌ ์ž…๋ ฅ ๋ฐ์ดํ„ฐ๋กœ ์‚ฌ์šฉ๋œ๋‹ค. ์‹คํ—˜ ์ง„ํ–‰ํ•˜๊ธฐ ์œ„ํ•ด GoogleNet์€ ์ตœ์ ํ™” ํ•จ์ˆ˜ Adam, ์ดˆ๊ธฐ ํ•™์Šต๋ฅ  1e-4, ์—ํฌํฌ 30, ๋ฏธ๋‹ˆ๋ฐฐ์น˜์‚ฌ์ด์ฆˆ 64๋กœ ์„ค์ •ํ•˜์˜€๋‹ค. ResNet-101์€ ์ตœ์ ํ™” ํ•จ์ˆ˜ Adam, ์ดˆ๊ธฐ ํ•™์Šต๋ฅ  1e-4, ์—ํฌํฌ 20, ๋ฏธ๋‹ˆ๋ฐฐ์น˜์‚ฌ์ด์ฆˆ 32๋กœ ์„ค์ •ํ•˜์˜€๋‹ค.

ํ‘œ 1์€ PTB ๋ฐ์ดํ„ฐ๋ฒ ์ด์Šค์˜ 2์ฐจ์› ํ•ฉ์„ฑ๊ณฑ ์‹ ๊ฒฝ๋ง์„ ์ด์šฉํ•œ ๊ฐœ์ธ์‹๋ณ„ ์„ฑ๋Šฅ ๊ฒฐ๊ณผ์ด๋‹ค. 2๊ฐœ์˜ ์ „์ดํ•™์Šต ๋ชจ๋ธ์„ ์‚ฌ์šฉํ•œ ๊ฒฐ๊ณผ GoogleNet์—์„œ STFT๊ฐ€ 98.04%๋กœ ๊ฐ€์žฅ ๋†’์€ ์„ฑ๋Šฅ์„ ๋‚˜ํƒ€๋‚ด๊ณ , FSST๋Š” 97.32%, WSST๋Š” 97.50%๋กœ ์„ฑ๋Šฅ์„ ํ™•์ธํ•  ์ˆ˜ ์žˆ๋‹ค. ๋˜ํ•œ ResNet-101์€ FSST๊ฐ€ 98.47%๋กœ ๊ฐ€์žฅ ์šฐ์ˆ˜ํ•œ ์„ฑ๋Šฅ์„ ๋‚˜ํƒ€๋‚ด๊ณ , STFT๋Š” 98.44%, WSST๊ฐ€ 98.29%์˜ ์„ฑ๋Šฅ์„ ํ™•์ธํ•  ์ˆ˜ ์žˆ๋‹ค. ์‹ฌ์ „๋„ ์‹ ํ˜ธ๋ฅผ 3๊ฐœ์˜ ๋ณ€ํ™˜ ๋ฐฉ๋ฒ•์„ ํ†ตํ•ด 2์ฐจ์› ์‹œ๊ฐ„-์ฃผํŒŒ์ˆ˜ ์˜์—ญ์œผ๋กœ ๋ณ€ํ™˜ํ•˜์—ฌ GoogleNet๊ณผ ResNet-101๋ฅผ ์‚ฌ์šฉํ•œ ๊ฒฐ๊ณผ๋Š” ๋ชจ๋‘ ์šฐ์ˆ˜ํ•˜์ง€๋งŒ ResNet-101์„ ์‚ฌ์šฉํ–ˆ์„ ๊ฒฝ์šฐ ๋” ์šฐ์ˆ˜ํ•œ ๊ฒƒ์„ ํ™•์ธํ•  ์ˆ˜ ์žˆ๋‹ค.

ํ‘œ 1 PTB ๋ฐ์ดํ„ฐ๋ฒ ์ด์Šค์˜ 2์ฐจ์› ํ•ฉ์„ฑ๊ณฑ ์‹ ๊ฒฝ๋ง๊ธฐ๋ฐ˜ ์‹คํ—˜ ๊ฒฐ๊ณผ

Table 1 2D Convolutional Neural Network Based Experimental Results of PTB Database

์‹œ๊ฐ„-์ฃผํŒŒ์ˆ˜ ํ‘œํ˜„

2D-CNN ๊ฒ€์ฆ ์ •ํ™•๋„

GoogleNet

STFT

98.04%

FSST

97.32%

WSST

97.50%

ResNet-101

STFT

98.44%

FSST

98.47%

WSST

98.29%

6. ๊ฒฐ ๋ก 

์‹ฌ์ „๋„๋ฅผ ์ด์šฉํ•œ ๊ฐœ์ธ์‹๋ณ„์€ ์‚ฌ์šฉ์ž๊ฐ€ ์‹ฌ์ „๋„ ์‹ ํ˜ธ๋ฅผ ๋“ฑ๋กํ•˜์—ฌ ์ €์žฅ๋œ ์‹ฌ์ „๋„ ์‹ ํ˜ธ์™€ ๋น„๊ตํ•˜์—ฌ ์‚ฌ์šฉ์ž๋ฅผ ์ธ์‹ํ•˜๋Š” ๊ธฐ์ˆ ์ด๋‹ค. ์‹ฌ์ „๋„๋Š” ์‹ฌ์žฅ์˜ ํฌ๊ธฐ, ์œ„์น˜, ๋‚˜์ด ๋ฐ ์„ฑ๋ณ„ ๋“ฑ์— ๋”ฐ๋ผ ๋‹ค๋ฅด๊ฒŒ ๋‚˜ํƒ€๋‚˜๋Š” ์‹ ํ˜ธ๋กœ ์‹ฌ์ „๋„์˜ ํŠน์„ฑ์„ ํ†ตํ•ด 90% ์ด์ƒ์ธ ๋†’์€ ํ™•๋ฅ ๋กœ ๊ฐœ์ธ์„ ๋ถ„๋ฅ˜ํ•  ์ˆ˜ ์žˆ๋‹ค. ๋˜ํ•œ ์‹ฌ์ „๋„๋ฅผ ์ด์šฉํ•˜์—ฌ ๊ฐœ์ธ์‹๋ณ„๋ฟ๋งŒ ์•„๋‹ˆ๋ผ ์˜๋ฃŒ ๋ถ„์•ผ๋กœ ์‹ฌ์žฅ ์งˆ๋ณ‘๋“ค์„ ์˜ˆ์ธกํ•˜๊ณ  ์ง„๋‹จํ•  ์ˆ˜ ์žˆ์œผ๋ฉฐ, ์‚ฌ์šฉ์ž์˜ ์‹ฌ์žฅ ์ƒํƒœ๋ฅผ ์›๊ฒฉ์œผ๋กœ ํ™•์ธ์ด ๊ฐ€๋Šฅํ•œ ๊ธฐ์ˆ ์„ ๊ธฐ๋Œ€ํ•  ์ˆ˜ ์žˆ๋‹ค.

๋ณธ ๋…ผ๋ฌธ์—์„œ๋Š” ์‹ฌ์ „๋„๊ธฐ๋ฐ˜์œผ๋กœ 2์ฐจ์› ํ•ฉ์„ฑ๊ณฑ ์‹ ๊ฒฝ๋ง์„ ์ด์šฉํ•˜์—ฌ ๊ฐœ์ธ์‹๋ณ„์„ ์ง„ํ–‰ํ–ˆ๋‹ค. ์ด๋ฅผ ์œ„ํ•ด ๋‹ค์–‘ํ•œ ์žก์Œ๋“ค๋กœ ์ธํ•œ ์™œ๊ณก๋œ ์‹ ํ˜ธ๋“ค์„ ํ‰๊ท  ์ด๋™ ํ•„ํ„ฐ, ์ €์—ญ ํ†ต๊ณผ ํ•„ํ„ฐ, ๊ณ ์—ญ ํ†ต๊ณผ ํ•„ํ„ฐ๋ฅผ ์ ์šฉํ•˜๊ณ  ๊ธฐ์ €์„ ์„ ์˜์ ์œผ๋กœ ๋งž์ถฐ ์™œ๊ณก๋œ ์‹ ํ˜ธ๋“ค์˜ ์žก์Œ์„ ์ œ๊ฑฐํ•˜์˜€๋‹ค. ๋˜ํ•œ ์‹ฌ์ „๋„ ์‹ ํ˜ธ์˜ ํŠน์ง•์„ ์ถ”์ถœํ•˜๊ธฐ ์œ„ํ•ด R-peak๋ฅผ ๊ฒ€์ถœํ•˜์—ฌ, R-peak๋ฅผ ๊ธฐ์ค€์œผ๋กœ ์ผ์ •ํ•œ ๊ฐ„๊ฒฉ์— ๋”ฐ๋ผ ์‹ ํ˜ธ๋ฅผ ํ•œ ์ฃผ๊ธฐ๋กœ ๋ถ„ํ• ํ•˜์˜€๋‹ค. ๋ถ„ํ• ๋œ 1์ฐจ์› ์‹ฌ์ „๋„ ์‹ ํ˜ธ๋Š” 3๊ฐ€์ง€์˜ ์‹œ๊ฐ„-์ฃผํŒŒ์ˆ˜ ๋ณ€ํ™˜ ๋ฐฉ๋ฒ• STFT, FSST, WSST์— ์˜ํ•ด ์ „์ดํ•™์Šต์ธ GoogleNet, ResNet-101์„ ์‚ฌ์šฉํ•˜์˜€๋‹ค. PTB ๋ฐ์ดํ„ฐ๋ฒ ์ด์Šค๋ฅผ ์ด์šฉํ•˜์—ฌ 2์ฐจ์› ํ•ฉ์„ฑ๊ณฑ ์‹ ๊ฒฝ๋ง๊ธฐ๋ฐ˜ ๊ฐœ์ธ์‹๋ณ„ ์„ฑ๋Šฅ ๊ฒฐ๊ณผ๋Š” GoogleNet์—์„œ STFT๊ฐ€ 98.04%๋กœ ๊ฐ€์žฅ ๋†’์€ ์„ฑ๋Šฅ์„ ๋‚˜ํƒ€๋‚ด๊ณ , ResNet-101์—์„œ FSST๊ฐ€ 98.47%๋กœ ๊ฐ€์žฅ ์šฐ์ˆ˜ํ•œ ์„ฑ๋Šฅ์„ ํ™•์ธํ•  ์ˆ˜ ์žˆ๋‹ค. ํ–ฅํ›„์—๋Š” ์‹ฌ์ „๋„๊ธฐ๋ฐ˜ ๊ฐœ์ธ์‹๋ณ„๋ฟ๋งŒ ์•„๋‹ˆ๋ผ ์‹ฌ์ „๋„๋ฅผ ์ด์šฉํ•˜์—ฌ ๊ฐ์ •์ธ์‹ ์—ฐ๊ตฌ๋ฅผ ์ง„ํ–‰ํ•  ๊ณ„ํš์ด๋‹ค.

Acknowledgements

๋ณธ ๋…ผ๋ฌธ์€ 2017๋…„๋„ ์ •๋ถ€(๊ต์œก๋ถ€)์˜ ์žฌ์›์œผ๋กœ ํ•œ๊ตญ์—ฐ๊ตฌ์žฌ๋‹จ์˜ ์ง€์›์„ ๋ฐ›์•„ ์ˆ˜ํ–‰๋œ ๊ธฐ์ดˆ์—ฐ๊ตฌ์‚ฌ์—…์ž„ (No.2017R1A6A1A03015496)

References

1 
D. Jeong, 2020, Artificial intelligence security threat, crime, and forensics: Taxonomy and open issues, IEEE Access, Vol. 8, pp. 184560-184574DOI
2 
J. W. Kim, P. K. Rhee, 2018, Image Recognition based on Adaptive Deep Learning, The Journal of the Institute of Internet, Broadcasting and Communication, Vol. 18, No. 1, pp. 113-117DOI
3 
L. Lu, J. Mao, W. Wang, G. Ding, Z. Zhang, Aug. 2020, A Study of Personal Recognition Method Based on EMG Signal, IEEE Transactions on Biomedical Circuits and Systems, Vol. 14, No. 4, pp. 681-691DOI
4 
H. S. Sin, C. Y. Hahm, N. K. Kim, M. K. Kim, S. H. Lee, Y. S. Kim, 2014, Trends of Emotional Information & Communication Technology, Electronics and Telecommunications Trends, Vol. 29, No. 5, pp. 30-39URL
5 
A. Barros, D. Rosรกrio, P. Resque, E. Cerqueira, 2019, Heart of IoT: ECG as biometric sign for authentication and identification, 2019 15th International Wireless Communications & Mobile Computing Conference (IWCMC), pp. 307-312DOI
6 
X. Jiang, et al., 2021, Cancelable HD-sEMG-Based Biometrics for Cross-Application Discrepant Personal Identification, IEEE Journal of Biomedical, Health Informatics, Vol. 25, No. 4, pp. 1070-1079DOI
7 
K. Rabuzin, M. Baca, M. Sajko, 2006, E-learning: Biometrics as a Security Factor, International Multi- Conference on Computing in the Global Information Technology, pp. 64-64DOI
8 
M. Ingale, R. Cordeiro, S. Thentu, Y. Park, N. Karimian, 2020, ECG biometric authentication: A comparative analysis, IEEE Access, Vol. 8, pp. 117853-117866DOI
9 
Q. Zhang, D. Zhou, X. Zeng, 2017, HeartID: A multiresolution convolutional neural network for ECG-Based biometric human identification in smart health applications, IEEE Access, Vol. 5, pp. 11805-11816DOI
10 
R. Bousseljot, D. Kreiseler, A. Schnabel, 1995., Nutzung der EKG-Signaldatenbank CARDIODAT der PTB รผber das Internet, Biomedizinische Technik, pp. 317DOI
11 
G. B. Moody, R. G. Mark., 2001, The impact of the MIT-BIH Arrhythmia Database, IEEE Engineering in Medicine and Biology Magazine, Vol. 20, No. 3, pp. 45-50DOI
12 
T. S. Lugovaya, 2005, Biometric human identification based on electrocardiogramGoogle Search
13 
S. A. Israel, J. M.Irvine, A. Cheng, M. D. Wiederhold, B. K. Wiederhold, 2005, ECG to identify individuals, Pattern Recognition, Vol. 38, No. 1, pp. 113-142DOI
14 
M. Jahiruzzaman, A. B. M. A. Hossain, 2015, ECG based biometric human identification using chaotic encryption, 2015 International Conference on Electrical Engineering and Information Communication Technology (ICEEICT), pp. 1-5DOI
15 
S. S. Abdeldayem, T. Bourlai, 2020, A novel approach for ECG-Based human identification using spectral correlation and deep learning, IEEE Transactions on Biometrics, Behavior, and Identity Science, Vol. 2, No. 1, pp. 1-14DOI
16 
R. D. Labati, E. Muรฑoz, V. Piuri, R. Sassi, F. Scotti, 2019, Deep-ECG: Convolutional neural networks for ECG biometric recognition, Pattern Recognition Letters, Vol. 126, pp. 78-85DOI
17 
Y. H. Byeon, K. C. Kwak, 2019, Pre-Configured Deep Convolutional Neural Networks with Various Time- Frequency Representations for Biometrics from ECG Signals, Applied Sciences, Vol. 9, No. 22, pp. 2076-3417DOI
18 
L. Biel, O. Pettersson, L. Philipson, P. Wide, 2001, ECG analysis: a new approach in human identification, IEEE Transactions on Instrumentation and Measurement, Vol. 50, No. 3, pp. 808-812DOI
19 
D. Jyotishi, S. Dandapat, 2020, An LSTM-Based Model for Person Identification Using ECG Signal, IEEE Sensors Letters, Vol. 4, No. 8, pp. 1-4DOI
20 
J. S. Kim, S. G. Kim, S. B. Pan, 2020, Personal recognition using convolutional nearal network with ECG coupling image, Journal of Ambient Intelligence and Humanized Computing, Vol. 11, pp. 1923-1932DOI
21 
M. Hammad, P. Pล‚awiak, K. Wang, U. R. Acharya, 2021, ResNet-Attention model for human authen ticationusing ECG signals, Expert Systems, Vol. 38, No. 6DOI
22 
Y. H. Byeon, S. B. Pan, K. C. Kwak, 2019., Intelligent deep models based on scalograms of electrocardiogram signals for biometrics, Sensors, Vol. 19, No. 4DOI
23 
K. El-Shennawy, 2014, Communication theory and signal processing for transform coding, Bentham Science PublishersGoogle Search
24 
T. Oberlin, S. Meignen, V. Perrier, 2014, The fourier-based synchrosqueezing transform, 2014 IEEE International Conference on Acoustic,s Speech and Signal Processing (ICASSP), pp. 315-319DOI
25 
A. Kumar, C. P. Gandhi, Y. Zhou, G. Vashishtha, R. Kumar, J. Xiang, 2020, Improved CNN for the diagnosis of engine defects of 2-wheeler vehicle using wavelet synchro-squeezed transform (WSST), Knowledge-Based Systems, Vol. 208DOI
26 
A. Lumini, L. Nanni, 2019, Deep learning and transfer learning features for plankton classification, Ecological Informatics, Vol. 51, pp. 33-43DOI
27 
K. He, X. Zhang, S. Ren, J. Sun, 2016, Deep residual learning for image recognition, 2016 IEEE Conference on Computer Vision and Pattern Recognition, pp. 770-778URL

์ €์ž์†Œ๊ฐœ

์ด์ง„์•„(Jin-A Lee)
../../Resources/kiee/KIEEP.2022.71.1.54/au1.png

2020๋…„ 2์›” : ์กฐ์„ ๋Œ€ํ•™๊ต ์ „์ž๊ณตํ•™๊ณผ(ํ•™์‚ฌ)

2020๋…„ 3์›”~ํ˜„์žฌ : ์กฐ์„ ๋Œ€ํ•™๊ต ์ „์ž๊ณตํ•™๊ณผ ์„์‚ฌ๊ณผ์ •

๊ด€์‹ฌ๋ถ„์•ผ : ์˜์ƒ์ฒ˜๋ฆฌ, ๋ฐ”์ด์˜ค์ธ์‹

๊ณฝ๊ทผ์ฐฝ(Keun-Chang Kwak)
../../Resources/kiee/KIEEP.2022.71.1.54/au2.png

2002๋…„: ์ถฉ๋ถ๋Œ€ํ•™๊ต ์ „๊ธฐ๊ณตํ•™๊ณผ ๋ฐ•์‚ฌ ์กธ์—…

2003๋…„~2005๋…„: ์บ๋‚˜๋‹ค ์•จ๋ฒ„ํƒ€๋Œ€ํ•™๊ต ์ „๊ธฐ ๋ฐ ์ปดํ“จํ„ฐ ๊ณตํ•™๊ณผ, ๋ฐ•์‚ฌํ›„๊ณผ์ •

2005๋…„~2007๋…„: ํ•œ๊ตญ์ „์žํ†ต์‹ ์—ฐ๊ตฌ์› ์ง€๋Šฅํ˜•๋กœ๋ด‡์—ฐ๊ตฌ๋‹จ ์„ ์ž„์—ฐ๊ตฌ์›

2014๋…„~2015๋…„: ๋ฏธ๊ตญ ์บ˜๋ฆฌํฌ๋‹ˆ์•„์ฃผ๋ฆฝ๋Œ€ํ•™๊ต ํ”Œ๋ŸฌํŠผ, ๋ฐฉ๋ฌธ๊ต์ˆ˜

2007๋…„~ํ˜„์žฌ: ์กฐ์„ ๋Œ€ํ•™๊ต ์ „์ž๊ณตํ•™๋ถ€ ๊ต์ˆ˜

๊ด€์‹ฌ๋ถ„์•ผ: ๊ณ„์‚ฐ์ง€๋Šฅ, ์ธ๊ฐ„-๋กœ๋ด‡์ƒํ˜ธ์ž‘์šฉ, ๋ฐ”์ด์˜ค์ธ์‹