对现代汉语“X的O”句的考察

공연 중국어문연구회 2019 中國語文論叢 Vol.0 No.95

In this paper, we argue that ‘shi...de’ construction in Modern Chinese should be regarded as a combination of ‘Shi’ and ‘De construction’. According to the grammatical position of ‘De’, we divided ‘De construction’ into two types. One is the ‘X De O’ construction, in which the grammatical position of ‘De’ is between predicate and object. The other is the ‘X De’ construction, in which the grammatical position of ‘De’ is in the end. This paper mainly focused on the syntactic features of predicate and object in the ‘X De O’ construction. We argue that the syntactic features are also syntactic conditions for the grammaticalization of structural particle ‘De’ into aspectual particle ‘De’. According to Cognitive Grammar, ‘De’ can be regarded as a grounding element. In ‘X De O’ phase, structural particle ‘De’ is a grounding element related to the grounding of a noun phrase and in ‘X De O’ construction, aspectual particle ‘De’ is also a grounding element which related to the grounding of a clause. The aspectual particle ‘De’ connects event time with speech time, and means event time is before speech time. Based on the theory of cognitive grammar, this paper also reveals the cognitive mechanism of the transfer of De’s function from spatial domain to time domain.

Investigation of characteristic values in TDR waveform using SHapley Additive exPlanations (SHAP) for dielectric constant estimation during curing time

Hyung-Koo Yoon,Won-Taek Hong,WooJin Han,Yong-Hoon Byun 국제구조공학회 2024 Smart Structures and Systems, An International Jou Vol.34 No.1

As materials cure, the internal electrical flow changes, leading to variations in the dielectric constant over time. This study aims to assess the impact of voltage values extracted from time domain reflectometry (TDR) waveforms, measured during the curing of materials, on predicting the dielectric constant. The experiments are conducted over a curing period ranging from 60 to 8640 minutes, with 30 TDR trials. From the measured waveforms, values of V₀, V₁, V₂, V_f, and Δt are deduced. Additionally, curing time is included as an input variable. Groups A and B are distinguished based on the presence or absence of Δt, indicating a physical relationship between Δt and the dielectric constant. The dielectric constant is set as the output variable. The SHapley Additive exPlanations (SHAP) algorithm is applied to the compiled data. The results indicate that Δt and V₁ are the most influential input variables in both Group-A and Group-B. The study also presents the distribution of SHAP values and interacts SHAP values to infer the interrelationships among the input variables. To validate the reliability of these findings, the partial dependence (PD) algorithm is applied to estimate the marginal effects of each input variable, with outcomes closely aligning with those of the SHAP algorithm. This research suggests that understanding the contributions and proportional relationships of each input variable can aid in interpreting the relationships among various material properties.

시간-주파수 영역 특성 응답을 이용한 매설 배관의 충격 손상 위치 표정에 대한 실험적 검증

이선호,박춘수,윤동진 한국비파괴검사학회 2023 한국비파괴검사학회지 Vol.43 No.4

Damage to buried pipelines resulting from third-party interference has garnered attention because it can lead to significant leaks owing to ruptures. Given that such incidents often occur suddenly owing to third-party actions, the importance of buried pipeline monitoring becomes paramount for prevention and early response. Acoustic emission localization have been recognized as an effective method for detecting damage sources in structures with restricted access. In this study, we propose a signal processing method based on the feature response in the time-frequency domain to apply acoustic emission localization in noisy environments. We compared the results of the feature response through simulation and experimentally verified its application on in-service buried pipelines. The pipeline is 564 m long and has a diameter of 1200 mm. By attaching vibration accelerometers to both ends, we confirmed the ability to detect impact damages. By observing the feature response, we derived the optimal frequency band without manual frequency analysis. Therefore, we could automatically enhance the signal-to-noise ratio under various noise conditions. Our study yielded an empirical impact damage detection in the in-service pipelines that was in good agreement with the impact location. 타공사에 의한 매설 배관 파손 사고는 파열로 인한 다량의 누출을 동반하여 사회적 문제로 주목받고 있다. 이러한 사고는 제 3 자에 의해 갑작스럽게 발생하는 관계로 예방과 조기 대응을 위해서는 매설 배관 모니터링 기술이 중요하다. 음향방출 위치 표정 기법은 접근이 제한되는 구조물을 대상으로 손상원 검출에 효과적인 방법으로 주목받아 왔다. 본 연구에서는 잡음 환경에서도 음향방출 위치 표정 기법을 적용하기위해 시간-주파수 영역의 특성 응답을 이용한 신호처리 기법을 제안한다. 시뮬레이션을 통한 특성 응답 결과를 비교하였으며, 이를 실제 운용중인 매설 배관에 적용하여 실험적으로 검증하였다. 대상 배관의 관경은1,200 mm이며, 길이 564 m에서 진동 가속도계를 부착하여 충격 손상의 검출이 가능함을 확인하였다. 특히, 특성 응답 관찰을 통해 인위적인 주파수 분석 없이 최적 주파수 대역을 도출하였다. 그 결과, 다양한 잡음조건에서 자동적으로 신호 대 잡음비 향상이 가능하였으며, 실제 충격 위치와 잘 일치함을 확인하였다.

Comparison of wavelet-based decomposition and empirical mode decomposition of electrohysterogram signals for preterm birth classification

Suparerk Janjarasjitt 한국전자통신연구원 2022 ETRI Journal Vol.44 No.5

Signal decomposition is a computational technique that dissects a signal into its constituent components, providing supplementary information. In this study, the capability of two common signal decomposition techniques, including wavelet-based and empirical mode decomposition, on preterm birth classification was investigated. Ten time-domain features were extracted from the constituent components of electrohysterogram (EHG) signals, including EHG subbands and EHG intrinsic mode functions, and employed for preterm birth classification. Preterm birth classification and anticipation are crucial tasks that can help reduce preterm birth complications. The computational results show that the preterm birth classification obtained using wavelet-based decomposition is superior. This, therefore, implies that EHG subbands decomposed through wavelet-based decomposition provide more applicable information for preterm birth classification. Furthermore, an accuracy of 0.9776 and a specificity of 0.9978, the best performance on preterm birth classificationamong state-of-the-art signal processing techniques, were obtained using the time-domain features of EHG subbands.

Characterizations on Knee Movement Estimation from Surface EMG Using Composited Approaches

Hui-Bin Li,Zhong Li,Xiao-Rong Guan 대한전기학회 2024 Journal of Electrical Engineering & Technology Vol.19 No.1

This paper focuses on the relationship between surface electromyography (sEMG) signals and knee joint movement which would be applied in the extra assistant robotic legs (EARL). The characterizations of the sEMG data group for movement estimation were analyzed depending on the composited approaches, which were analyzed in situations of diferent sample sizes, diferent muscle groups, diferent time-domain features, and diferent algorithms. According to the analysis of knee movement estimation, it can be found that (1) Increasing the sample size, and the estimation accuracy becomes better. (2) The vastus lateralis and gastrocnemius lateralis have larger activation than the vastus medialis and gastrocnemius medialis, but the estimation accuracy is worse. (3) The estimation accuracy of time-domain features is related to the complexity of the feature group, the range of average AR2 is 0.86 to 0.92. (4) The composited algorithms are better than a single back propagation neural network (BPNN), and principal component analysis is ftter than independent component analysis in conjunction with BPNN for estimation. This study reveals the hind factors that people cannot square up while extracting and processing the sEMG signals. This study’s results can help build new training data to improve the estimation accuracy of knee movement for sEMG.

A Continuous Abnormal Speech Detection Method Based on Time Domain features Weighted

He Jun,Ji-chen Yang,Qing-hua Zhang,Guo-xi Sun,Jian-bing Xiong 보안공학연구지원센터 2014 International Journal of Control and Automation Vol.7 No.12

In this brief, a novel pathological continuous speech detection method based on time domain features weighted. First, different optimal threshold for time domain features, including zero crossing ratio, short-time energy and autocorrelation, are obtained from training speech data. Second, a difference evaluation technique is proposed, and with it, the difference of the same time domain feature selected from testing speech data and training speech data were obtained. Finally, to distinguish a given speech well, a novel weighting method based on difference evaluation for each kinds of time domain is employed, respectively. Experiments were conducted on the pathological speech database to prove the power and effectiveness of the proposed method. Results obtained shown that this method outperforms other early proposed time domain feature method, creating a more reliable technique for pathological continuous speech detection.

Human Action Recognition Based on 3D Convolutional Neural Network from Hybrid Feature

Tingting Wu,이응주 한국멀티미디어학회 2019 멀티미디어학회논문지 Vol.22 No.12

3D convolution is to stack multiple consecutive frames to form a cube, and then apply the 3D convolution kernel in the cube. In this structure, each feature map of the convolutional layer is connected to multiple adjacent sequential frames in the previous layer, thus capturing the motion information. However, due to the changes of pedestrian posture, motion and position, the convolution at the same place is inappropriate, and when the 3D convolution kernel is convoluted in the time domain, only time domain features of three consecutive frames can be extracted, which is not a good enough to get action information. This paper proposes an action recognition method based on feature fusion of 3D convolutional neural network. Based on the VGG16 network model, sending a pre-acquired optical flow image for learning, then get the time domain features, and then the feature of the time domain is extracted from the features extracted by the 3D convolutional neural network. Finally, the behavior classification is done by the SVM classifier.

시간 및 주파수 영역 특징 기반 양방향 LSTM 모델을 이용한 음성감정인식 기법

민동진,김덕환 한국차세대컴퓨팅학회 2023 한국차세대컴퓨팅학회 논문지 Vol.19 No.6

현재 음성에서 감정 상태를 이해하기 위해 강조, 음높이 변화 및 맥락과 같은 감정적 특징을 인식하는 다양한 음성감정인식(Speech Emotion Recognition)연구가 활성화 되고 있다. 본 논문은 음성 데이터에서 감정적 단서를 찾기 위해 시간 영역과 주파수 영역의 다양한 특징을 추출하고 강조, 음높이 변화를 통해 감정을 인식하는 신경망 모델을 개발하고자 한다. 부족한 데이터 셋으로 인한 과대적합을 예방하기 위해 컴퓨터 비전 분야에서 활용되는 데이터 증강 기법을 적용하였다. 또한 음성 데이터의 전처리를 수행하고 시간 영역에서 제로 크로싱 비율 (ZCR)과 RMS (Root mean square) 에너지 특징을 추출하였고 주파수 영역에서 멜 주파수 켑스트랄 계수 (MFCCs), 주파수 대역폭(Spectral Bandwidth), 주파수 중심 (Spectral Centroid), 주파수 롤오프의 최대, 최소 값(Spectral Rolloff Max and Min) 같은 스펙트럴 특징을 추출하였다. 과거와 미래의 정보를 저장할 수 있어 불연속적인 음성 데이터의 과거와 미래의 정보를 효과적으로 학습할 수 있는 양방향 LSTM 신경망 모델을 제안하였으며, 8가지 감정표현(중립, 차분함, 행복, 슬픔, 분노, 두려움, 혐오, 놀람)이 포함되어 있는 RAVDESS와 7가지 감정표현(중립, 행복, 슬픔, 분노, 두려움, 혐오, 놀람)이 포함되어 있는 TESS 데이터 셋에 대하여 각각 99.21%, 98.24%의 정확도를 확인하였다. 향후 연구에서는 음성 감정 인식 분야의 주요 어려움 중 하나인 적은 데이터 셋 문제를 메타러닝 기법을 활용하여 해결할 계획이다. Currently, various studies are actively underway to understand emotional states in speech, recognizing emotional features such as emphasis, pitch changes, and context. This paper aims to develop a neural network model that extracts various features from the time and frequency domains in speech data and recognizes emotions through emphasis and pitch changes. To prevent underfitting and overfitting due to limited datasets, data augmentation techniques commonly used in computer vision were applied. Additionally, preprocessing of the speech data was performed and the time domain features such as Zero Crossing Rate (ZCR) and Root Mean Square (RMS) energy were extracted. Similarly, the frequency domain spectral features such as Mel-frequency cepstral coefficients (MFCCs), Spectral Bandwidth, Spectral Centroid, and Spectral Rolloff Max and Min were extracted. To effectively handle discontinuous speech data by storing past and future information, a bidirectional LSTM model was proposed. On the RAVDESS dataset, which includes eight types of emotional expressions (neutral, calm, happy, sad, angry, fearful, disgusted, surprised), and the TESS dataset, which includes seven types of emotional expressions (neutral, happy, sad, angry, fearful, disgusted, surprised), we confirmed accuracies of 99.21% and 98.24%, respectively. In future research, we plan to address one of the major challenges in the field of speech emotion recognition, the scarcity of datasets, by utilizing meta-learning techniques.

뇌파 신호 기반 스트레스 상태 분류

강준수,장길진,이민호 한국인터넷방송통신학회 2016 한국인터넷방송통신학회 논문지 Vol.16 No.3

일상생활에서 인간은 끊임없이 스트레스를 받으며 살아간다. 스트레스는 삶의 질과 밀접하게 연관이 있으며, 건강한 삶은 스트레스에 적절하게 대처하며 살아가는 삶이다. 스트레스는 호르몬 분비에 영향을 주며, 호르몬 분비의 변화는 뇌 신호 및 생체 신호에 영향을 준다. 이를 바탕으로, 본 논문은 스트레스와 뇌파 신호와의 관련성을 확인하였으며, 더 나아가 뇌파 신호 기반 정량적 스트레스 지수를 찾아보았다. 사용한 뇌파 장비는 32채널 유선 EEG 장비이며, 상업용 2채널(FP1, FP2) 뇌파 장비와의 비교를 위해, 상업용 뇌파 장비와 동일한 위치에 있는 2채널만 이용하여 데이터를 분석하였다. 뇌파의 주파수 특징점으로는 각 주파수 대역대의 파워 값, 주파수 대역대 파워 값들 간의 비율 및 차이 등을 테스트해 보았으며, 시간 특징점으로는 허스트 지수, 상관 지수, 리아프노프 지수 등을 테스트해 보았다. 총 6명의 피 실험자가 본 실험에 참여하였으며, 실험 과제로는 영어 지문이 사용되었다. 여러 특징점들 중 가 가장 좋은 테스트 성능을 보여줬으며, 테스트 데이터에 대하여 평균 70.8%의 스트레스 분류 정확도를 얻었다. 추후, 저가 상용 2채널 뇌파 장치를 이용해서 비슷한 결과가 나오는지 확인해 볼 예정이다. In daily life, humans get stress very often. Stress is one of the important factors of healthy life and closely related to the quality of life. Too much stress is known to cause hormone imbalance of our body, and it is observed by the brain and bio signals. Based on this, the relationship between brain signal and stress is explored, and brain signal based stress index is proposed in our work. In this study, an EEG measurement device with 32 channels is adopted. However, only two channels (FP1, FP2) are used to this study considering the applicability of the proposed method in real enveironment, and to compare it with the commercial 2 channel EEG device. Frequency domain features are power of each frequency bands, subtraction, addition, or division by each frequency bands. Features in time domain are hurst exponent, correlation dimension, lyapunov exponent, etc. Total 6 subjects are participated in this experiment with English sentence reading task given. Among several candidate features, shows the best test performance (70.8%). For future work, we will confirm the results is consistent in low price EEG device.

Physical Activity Recognition for Mobile Devices Based on Time Domain Features with Periodic Calculation

Mamatov Fahriddin,Min Gyu Song,Tran Thi My Hue,Jin Young Kim,Seung You Na 한국정보기술학회 2011 한국정보기술학회논문지 Vol.9 No.10