• ETRI Journal
• /
• 제39권3호
• /
• pp.398-405
• /
• 2017

In this paper, an artificial stereo extension method that creates stereophonic sound from a mono sound source is proposed. The proposed method first trains deep neural networks (DNNs) that model the nonlinear relationship between the dominant and residual signals of the stereo channel. In the training stage, the band-wise log spectral magnitude and unwrapped phase of both the dominant and residual signals are utilized to model the nonlinearities of each sub-band through deep architecture. From that point, stereo extension is conducted by estimating the residual signal that corresponds to the input mono channel signal with the trained DNN model in a sub-band domain. The performance of the proposed method was evaluated using a log spectral distortion (LSD) measure and multiple stimuli with a hidden reference and anchor (MUSHRA) test. The results showed that the proposed method provided a lower LSD and higher MUSHRA score than conventional methods that use hidden Markov models and DNN with full-band processing.

• ETRI Journal
• /
• 제42권5호
• /
• pp.658-668
• /
• 2020

In edge computing, most procedures, including data collection, data processing, and service provision, are handled at edge nodes and not in the central cloud. This decreases the processing burden on the central cloud, enabling fast responses to end-device service requests in addition to reducing bandwidth consumption. However, edge nodes have restricted computing, storage, and energy resources to support computation-intensive tasks such as processing deep neural network (DNN) inference. In this study, we analyze the effect of models with single and multiple local exits on DNN inference in an edge-computing environment. Our test results show that a single-exit model performs better with respect to the number of local exited samples, inference accuracy, and inference latency than a multi-exit model at all exit points. These results signify that higher accuracy can be achieved with less computation when a single-exit model is adopted. In edge computing infrastructure, it is therefore more efficient to adopt a DNN model with only one or a few exit points to provide a fast and reliable inference service.

• 대한의용생체공학회:의공학회지
• /
• 제40권2호
• /
• pp.48-54
• /
• 2019

This paper proposed a method of validation data augmentation for improving the grading accuracy of diabetic macular edema (DME) using deep learning. The data augmentation technique is basically applied in order to secure diversity of data by transforming one image to several images through random translation, rotation, scaling and reflection in preparation of input data of the deep neural network (DNN). In this paper, we apply this technique in the validation process of the trained DNN, and improve the grading accuracy by combining the classification results of the augmented images. To verify the effectiveness, 1,200 retinal images of Messidor dataset was divided into training and validation data at the ratio 7:3. By applying random augmentation to 359 validation data, $1.61{\pm}0.55%$ accuracy improvement was achieved in the case of six times augmentation (N=6). This simple method has shown that the accuracy can be improved in the N range from 2 to 6 with the correlation coefficient of 0.5667. Therefore, it is expected to help improve the diagnostic accuracy of DME with the grading information provided by the proposed DNN.

• Nuclear Engineering and Technology
• /
• 제51권3호
• /
• pp.723-730
• /
• 2019

Acquiring instrumentation signals generated from nuclear power plants (NPPs) is essential to maintain nuclear reactor integrity or to mitigate an abnormal state under normal operating conditions or severe accident circumstances. However, various safety-critical instrumentation signals from NPPs cannot be accurately measured on account of instrument degradation or failure under severe accident circumstances. Reactor vessel (RV) water level, which is an accident monitoring variable directly related to reactor cooling and prevention of core exposure, was predicted by applying a few signals to deep neural networks (DNNs) during severe accidents in NPPs. Signal data were obtained by simulating the postulated loss-of-coolant accidents at hot- and cold-legs, and steam generator tube rupture using modular accident analysis program code as actual NPP accidents rarely happen. To optimize the DNN model for RV water level prediction, a genetic algorithm was used to select the numbers of hidden layers and nodes. The proposed DNN model had a small root mean square error for RV water level prediction, and performed better than the cascaded fuzzy neural network model of the previous study. Consequently, the DNN model is considered to perform well enough to provide supporting information on the RV water level to operators.

• 한국멀티미디어학회논문지
• /
• 제22권9호
• /
• pp.1069-1079
• /
• 2019

In this study, we developed a $PM_{10}$ forecasting model using DNN and Membership Function, and improved the forecasting performance. The model predicts the $PM_{10}$ concentrations of the next 3 days in the Seoul area by using the weather and air quality observation data and forecast data. The best model(RM14)'s accuracy (82%, 76%, 69%) and false alarm rate(FAR:14%,33%,44%) are good. Probability of detection (POD: 79%, 50%, 53%), however, are not good performance. These are due to the lack of training data for high concentration $PM_{10}$ compared to low concentration. In addition, the model dose not reflect seasonal factors closely related to the generation of high concentration $PM_{10}$. To improve this, we propose Julian date membership function as inputs of the $PM_{10}$ forecasting model. The function express a given date in 12 factors to reflect seasonal characteristics closely related to high concentration $PM_{10}$. As a result, the accuracy (79%, 70%, 66%) and FAR (24%, 48%, 46%) are slightly reduced in performance, but the POD (79%, 75%, 71%) are up to 25% improved compared with those of the RM14 model. Hence, this shows that the proposed Julian forecast model is effective for high concentration $PM_{10}$ forecasts.

• 대한건축학회논문집:계획계
• /
• 제36권4호
• /
• pp.41-49
• /
• 2020

The purpose of this study was to investigate the use of the Deep Neural Networks(DNN) model to classify user's emotions, in particular Electroencephalography(EEG) toward Virtual-Reality(VR) based 3D design alternatives. Four different types of VR Space were constructed to measure a user's emotion and EEG was measured for each stimulus. In addition to the quantitative evaluation based on EEG data, a questionnaire was conducted to qualitatively check whether there is a difference between VR stimuli. As a result, there is a significant difference between plan types according to the normalized ranking method. Therefore, the value of the subjective questionnaire was used as labeling data and collected EEG data was used for a feature value in the DNN model. Google TensorFlow was used to build and train the model. The accuracy of the developed model was 98.9%, which is higher than in previous studies. This indicates that there is a possibility of VR and Fast Fourier Transform(FFT) processing would affect the accuracy of the model, which means that it is possible to classify a user's emotions toward VR based 3D design alternatives by measuring the EEG with this model.

• International Journal of Computer Science & Network Security
• /
• 제21권3호
• /
• pp.229-234
• /
• 2021

The explosive growth of video-based services is considered as the dominant contributor to Internet traffic. Hence it is very important for video service providers to meet the quality expectations of end-users. In the past, the Quality of Service (QoS) was the key performance of networks but it considers only the network performances (e.g., bandwidth, delay, packet loss rate) which fail to give an indication of the satisfaction of users. Therefore, Quality of Experience (QoE) may allow content servers to be smarter and more efficient. This work is motivated by the inherent relationship between the QoE and the QoS. We present a no-reference (NR) prediction model based on Deep Neural Network (DNN) to predict video QoE. The DNN-based model shows a high correlation between the objective QoE measurement and QoE prediction. The performance of the proposed model was also evaluated and compared with other types of neural network architectures, and three known machine learning methodologies, the performance comparison shows that the proposed model appears as a promising way to solve the problems.

• 전력전자학회논문지
• /
• 제26권1호
• /
• pp.1-8
• /
• 2021

For the safe and reliable operation of lithium-ion batteries in electric vehicles or energy storage systems, having accurate information of the battery, such as the state of charge (SOC), is essential. Many different techniques of battery SOC estimation have been developed, such as the Kalman filter. However, when this filter is applied to multiple batteries, it has difficulty maintaining the accuracy of the estimation over all cells owing to the difference in parameter values of each cell. The difference in the parameter of each cell may increase as the operation time accumulates due to aging. In this paper, a novel deep neural network (DNN)-based SOC estimation method for multi-cell application is proposed. In the proposed method, DNN is implemented to determine the nonlinear relationships of the voltage and current at different SOCs and temperatures. In the training, the voltage and current data obtained at different temperatures during charge/discharge cycles are used. After the comprehensive training with the data obtained from the cycle test with a cell, the resulting algorithm is applied to estimate the SOC of other cells. Experimental results show that the mean absolute error of the estimation is 1.213% at 25℃ with the proposed DNN-based SOC estimation method.

• 한국항공우주학회지
• /
• 제50권5호
• /
• pp.317-323
• /
• 2022

본 논문에서는 c-스파 단면을 갖는 복합재 로터 블레이드에 대해 co-rotational(CR) 이론 기반 비선형 쉘 요소를 사용하는 In-house code를 통해 고유진동수를 구하고, 이를 이용하여 블레이드의 진동특성을 예측하는 심층신경망 모델을 개발하였다. 심층신경망 모델은 무작위 두께 분포를 갖는 데이터와 스팬 방향으로 두께 감소 경향성을 보이는 데이터를 통해 심층신경망 모델의 정확성을 평가하였다.

• 한국수자원학회:학술대회논문집
• /
• 한국수자원학회 2020년도 학술발표회
• /
• pp.39-39
• /
• 2020

Low-flow 하천에서의 최저수위를 나타내는 지표이다. 일반적으로 유황곡선의 갈수량(Q355)를 대표적으로 사용한다. Low-flow는 물 공급 관리 및 계획, 관개용수, 생태계등 다양한 분야에 영향을 미친다. 이러한 Low-flow를 산정하기 위해서는 충분한 기간의 유량자료가 필요하다. 하지만 국토의 70%가 산지지형으로 구성되어 있는 우리나라의 경우 국가하천과 1급하천을 제외한 산지유역은 수위관측소가 부재하거나 결측으로 인해 자료가 충분하지 않아 Low-flow분석에 한계가 있다. 이에 과거에는 미계측지역의 갈수량을 예측하기 위해서 다중회귀분석, ARIMA 모형 등 다양한 기법을 사용하였지만, 최근들어 머신러닝 모형의 수요가 증가하고 있다. 이에 본 연구에서는 새로운 패러다임에 맞는 머신러닝 기법인 DNN기법을 사용하고자 한다. DNN기법은 ANN기법의 단점인 학습과정에서 최적 매개변수값을 찾기 어렵고, 학습시간이 느린 단점을 보완한 방법이다. 따라서 본연구에서는 머신러닝 기법인 DNN기법을 통해 미계측지역에 적용 가능한 지역화 Low-flow indices를 산정하고자 한다. 먼저, Low-flow에 영향을 미치는 인자들을 수집하고 인자들간의 상관분석, 다중공선성 분석을 통해 통계적으로 유의한 변수를 선정하여, 머신러닝 모형에 입력자료를 구축하였다. 또한 기존의 갈수량 예측기법인 다중회귀분석 결과와 비교하여 머신러닝 기법의 효용성을 검토하였다.