[논문]LSTM/RNN을 사용한 감정인식을 위한 스택 오토 인코더로 EEG 차원 감소

임창균

doi:10.13067/jkiecs.2020.15.4.717

초록
AI-Helper

감성 컴퓨팅은 인간의 상호 작용에서 중요한 역할을 하기 때문에 인간을 인식하는 인공 지능을 통해 감정을 이해하고 식별한다. 우울증, 자폐증, 주의력 결핍 과잉 행동 장애 및 게임 중독과 같은 정신 질환을 잘 이해함으로써 감정과 관련된 문제들을 잘 관리할 수 있을 것이다. 이러한 문제들을 해결하기 위해 감정 인식을 위한 다양한 연구가 수행되었는데 기계학습을 적용하는데 있어서는 알고리즘의 복잡성을 줄이고 정확도를 향상시키기 위한 노력이 필요하다. 본 논문에서는 이러한 노력중의 하나로 Stack AutoEncoder (SAE)를 이용하여 차원을 감소하는 방법과 Long-Short-Term-Memory/Recurrent Neural Networks (LSTM / RNN) 분류를 이용한 감성 분류에 대해 연구한 결과를 제시한다. 제안된 방법은 모델의 복잡성을 줄이고 분류기의 성능을 크게 향상시킨 결과를 가져왔다.

Abstract ▼ AI-Helper

Due to the important role played by emotion in human interaction, affective computing is dedicated in trying to understand and regulate emotion through human-aware artificial intelligence. By understanding, emotion mental diseases such as depression, autism, attention deficit hyperactivity disorder,...

Due to the important role played by emotion in human interaction, affective computing is dedicated in trying to understand and regulate emotion through human-aware artificial intelligence. By understanding, emotion mental diseases such as depression, autism, attention deficit hyperactivity disorder, and game addiction will be better managed as they are all associated with emotion. Various studies for emotion recognition have been conducted to solve these problems. In applying machine learning for the emotion recognition, the efforts to reduce the complexity of the algorithm and improve the accuracy are required. In this paper, we investigate emotion Electroencephalogram (EEG) feature reduction and classification using Stack AutoEncoder (SAE) and Long-Short-Term-Memory/Recurrent Neural Networks (LSTM/RNN) classification respectively. The proposed method reduced the complexity of the model and significantly enhance the performance of the classifiers.

주제어

표/그림 (12)

그림 Fig. 1 Block diagram of the proposed model
그림 Fig. 2 Protocol of the EEG experiment [7]
그림 Fig. 3 Process of EEG feature dimensional reduction
그림 Fig. 4 The structure of standard RNN and LSTM [2].
그림 Fig. 5 SAE model loss performance.
표 Table 1. Model hyperparameters for investigation
표 Table 2. SAE Model Training and Validation Loss
그림 Fig. 6 Model Accuracy/Loss with various number of feature reduction (a) 12 features (b) 22 features (c) 32 features.
그림 Fig. 7 Validation accuracy of various number of feature reduction
그림 Fig. 8 LSTM/RNN Model without SAE using 62 features
그림 Fig. 9 Models recognition performance on various emotions
그림 Fig. 10 Result evaluation against relevant methods

AI 본문요약
AI-Helper

* AI 자동 식별 결과로 적합하지 않은 문장이 있을 수 있으니, 이용에 유의하시기 바랍니다.

문제 정의

In this paper, we investigated the used of SAE for dimensional reduction in emotion recognition. We obtained the best model with 32 features.

제안 방법

In the section, we investigate the performance of the model against other models and works. First, we trained our LSTM/RNN model with the 62 features from the dataset without dimensional reduction. As shown in Fig.
In Pre-processing stage, three operations were conducted, which include down-sampling, EOG/EMG removal and bandpass filtering. The EEG data was first down sampled to the rate of 200Hz.
The details of each part are given below. In the proposed model, SAE was used for feature dimensional reduction in order to reduce the dimension of the collected EEG signals. LSTM was the main component that was used in the emotion timing model to recognize emotion using the EEG feature sequence based on the EEG features produced by SAE.
Despite the efforts in emotions recognition, emotion detection from nonstationary, EEG signals is still a challenging task as a sophisticated learning algorithm that can represent high-level abstraction is required. In this paper, we investigate SAE for feature dimensionality reduction and LSTM classifier for emotion recognition in order to reduce the complexity of features and obtain high accuracy.
In this work, we are focused on recognizing emotion, thus we cast the problem as multilabel classification. Given a series of observations x⁽¹⁾, x⁽²⁾ , x⁽³⁾ ,.
The emotion EEG dataset used for this study is SEED obtained from BCMI laboratory and published in [7]. The emotion EEG data were collected from 15 subjects who were exposed to emotional films with real-life scenarios capable of triggering positive, neutral and negative emotions. Fig.
The purpose of this work is to determine an encoder that allows us to effectively reduce the dimension of EEG data from 62 features. SAE is chosen for this task because it consists of multiple layers that can achieve better performance.

대상 데이터

The experiments were conducted using 7500 data, where 75% of the data was used for training while 25% was used for validation for both SAE and LSTM. Separate 1000 test data were used to evaluate the model. Several experiments were conducted to determine the configuration of the model.
The experiments were conducted using 7500 data, where 75% of the data was used for training while 25% was used for validation for both SAE and LSTM. Separate 1000 test data were used to evaluate the model.

이론/모형

The emotion EEG dataset used for this study is SEED obtained from BCMI laboratory and published in [7].

성능/효과

Comparing the validation accuracy attained by the models, 12 features performed poorly with an accuracy of 69.4%, 22 features gave good performance of 84.9% and 32 features gave the best accuracy of 87.5%. The results are presented in Fig.
Other techniques that we compare against include Logistic Regression (LR), Random Forest (RF) and Decision Tree (DT). From the result presented in Fig. 10, our model gave an outstanding overall recognition accuracy of 86.6%, followed by that of Zheng and Lu model with 86.1% accuracy. Xing, et al model obtained a mean accuracy of 77.
3%. Neutral emotion enjoys a relatively good recognition accuracy among the models with 62 features (with SAE reduction) recording the highest accuracy result of 91.4% while 12 features performed the least with 82.2% accuracy. Happy emotion enjoyed the highest recognition accuracy of 99.

참고문헌 (11)

S. Koelstra et al., "Deap: A database for emotion analysis; using physiological signals," IEEE transactions on affective computing, vol. 3, no. 1, 2011, pp. 18-31.
Yun-Seok Jang and Jae-Woong Han "Analysis of EEG Generated from Concentration by Visual Stimulus Task", JKIECS, vol. 9, no. 5, 2014, 589-594.
Z. Li, X. Tian, L. Shu, X. Xu, and B. Hu, "Emotion recognition from eeg using rasm and lstm," in International Conference on Internet Multimedia Computing and Service, Springer. 2017, pp. 310-318.
M. Soleymani, S. Asghari-Esfeden, Y. Fu, and M. Pantic, "Analysis of EEG signals and facial expressions for continuous emotion detection," IEEE Transactions on Affective Computing, vol. 7, no. 1, 2015, pp. 17-28.
X. Li, D. Song, P. Zhang, G. Yu, Y. Hou, and B. Hu, "Emotion recognition from multi-channel EEG data through Convolutional Recurrent Neural Network," in 2016 IEEE International Conference on Bioinformatics and Biomedicine (BIBM), 2016, pp. 352-359.
X. Xing, Z. Li, T. Xu, L. Shu, B. Hu, and X. Xu, "SAE+ LSTM: A New Framework for Emotion Recognition from Multi-Channel EEG," Frontiers in Neurorobotics, vol. 13, 2019, p. 1-14.
W.-L. Zheng and B.-L. Lu, "Investigating critical frequency bands and channels for EEG-based emotion recognition with deep neural networks," IEEE Transactions on Autonomous Mental Development, vol. 7, no. 3, 2015, pp. 162-175.

상세보기
H. Kim, K. Seok, and H. Sin, "Domestic radio waves propagate management and control systems investigate the system status," J. of the Korea Institute of Electronic Communication Sciences, vol. 11, no. 5, 2016, pp. 441-450.
H. Kim and K. Seok, "Domestic radio waves propagate management and control systems investigate the system status," J. of the Korea Institute of Electronic Communication Sciences, vol. 12, no. 1, 2017, pp. 1-8.
W. Choi and K. Seok, "Survey on ways to improve the system in preparation for changes in the radio management system," J. of the Korea Institute of Electronic Communication Sciences, vol. 13, no. 6, 2018, pp. 1145-1154.
J. Yang, K. Seok, and H. Sin, "Technological and Social Significance of the Revision of the Radio Law," J. of the Korea Institute of Electronic Communication Sciences, vol. 14, no. 4, 2019, pp. 627-636.

내보내기 구분	파일저장 인쇄 메일전송
구성항목	기본정보 상세정보 관리번호, 논문명, 저널/프로시딩명, 저자 , 발행년, 권, 호, 시작페이지, 끝페이지, 발행기관 관리번호, 논문명, 대등논문명, 저자 , 저널/프로시딩명, 발행기관, 발행년, 발행언어, 권, 호, 시작페이지, 끝페이지, ISBN, ISSN, 주제분야, 키워드, 초록(한글), 초록(영문), 저자(소속기관)
저장형식	Text(ASCII format) Excel format RefWorks Direct Export RIS format (for Reference Manager, ProCite, EndNote), Scholar's Aids, Mendeley
메일정보	받는사람 (필수) @ 보내는사람 (선택) @ 제목 내용 KISTI 검색결과 이메일 서비스
안내	총 건의 자료가 검색되었습니다. 다운받으실 자료의 인덱스를 입력하세요. (1-10,000) 검색결과의 순서대로 최대 10,000건 까지 다운로드가 가능합니다. 데이타가 많을 경우 속도가 느려질 수 있습니다.(최대 2~3분 소요) 다운로드 파일은 UTF-8 형태로 저장됩니다. 파일의 내용이 제대로 보이지 않을실 때는 웹브라우저 상단의 보기 -> 인코딩 -> 자동선택 여부를 확인하십시오. ~ Text(ASCII format) Excel format

연합인증

LSTM/RNN을 사용한 감정인식을 위한 스택 오토 인코더로 EEG 차원 감소
EEG Dimensional Reduction with Stack AutoEncoder for Emotional Recognition using LSTM/RNN 원문보기

초록
AI-Helper

Abstract ▼ AI-Helper

주제어

표/그림 (12)

표/그림 (12)

AI 본문요약
AI-Helper

문제 정의

제안 방법

대상 데이터

이론/모형

성능/효과

참고문헌 (11)

이 논문을 인용한 문헌

저자의 다른 논문 :

관련 콘텐츠

원문 보기

원문 URL 링크

오픈액세스(OA) 유형

이 논문과 함께 이용한 콘텐츠

AI-Helper ※ AI-Helper는 오픈소스 모델을 사용합니다.

선택된 텍스트

연합인증

LSTM/RNN을 사용한 감정인식을 위한 스택 오토 인코더로 EEG 차원 감소 EEG Dimensional Reduction with Stack AutoEncoder for Emotional Recognition using LSTM/RNN 원문보기

초록 용어보기논문에서 용어와 풀이말을 자동 추출한 결과로, 시범 서비스 중입니다. AI-Helper

Abstract ▼ AI-Helper

주제어

표/그림 (12) 모든 표/그림 보기

표/그림 (12) 슬라이드로 보기

AI 본문요약 엑셀 다운로드 AI-Helper

문제 정의

제안 방법

대상 데이터

이론/모형

성능/효과

참고문헌 (11)

이 논문을 인용한 문헌

저자의 다른 논문 :

임창균 (13)

관련 콘텐츠

원문 보기

원문 URL 링크

오픈액세스(OA) 유형

이 논문과 함께 이용한 콘텐츠

AI-Helper ※ AI-Helper는 오픈소스 모델을 사용합니다.

선택된 텍스트

LSTM/RNN을 사용한 감정인식을 위한 스택 오토 인코더로 EEG 차원 감소
EEG Dimensional Reduction with Stack AutoEncoder for Emotional Recognition using LSTM/RNN 원문보기

초록
AI-Helper

표/그림 (12)

표/그림 (12)

AI 본문요약
AI-Helper