• Journal of Korean Neurosurgical Society
• /
• v.60 no.4
• /
• pp.404-416
• /
• 2017

Objective : Functional and neural tissue recovery has been reported in many animal studies conducted with stem cells. However, the combined effect of cytokines and stem cells has not yet been adequately researched. Here, we analyzed the additive effects of granulocyte colony-stimulating factor (GCSF) on adipose-derived stem cells (ADSCs) infusion in the treatment of acute spinal cord injury (SCI) in rats. Methods : Four days after intrathecal infusion tubes implantation in Sprague-Dawley rats, SCI was induced with an infinite horizon impactor. In the Sham group (n=5), phosphate-buffered saline was injected 3, 7, and 14 days after SCI. GCSF, ADSCs, and ADSCs with GCSF were injected at the same time in the GCSF (n=8), ADSC (n=8), and ADSC+GCSF groups (n=7), respectively. Results : The ADSC and ADSC+GCSF groups, but not the GCSF group, showed significantly higher Basso-Beattie-Bresnahan scores than the Sham group during 8 weeks (p<0.01), but no significant difference between the ADSC and ADSC+GCSF groups. In the ladder rung test, all four groups were significantly different from each other, with the ADSC+GCSF group showing the best improvement (p<0.01). On immunofluorescent staining (GAP43, MAP2), western blotting (GAP43), and reverse transcription polymerase chain reaction (GAP43, nerve growth factor), the ADSC and ADSC+GCSF groups showed higher levels than the Sham and GCSF groups. Conclusion : Our analyses suggest that the combination of GCSF and ADSCs infusions in acute SCI in the rat does not have a significant additive effect. Hence, when combination agents for SCI stem cell therapy are considered, molecules other than GCSF, or modifications to the methodology, should be investigated.

• Journal of the Korea Society of Computer and Information
• /
• v.24 no.10
• /
• pp.33-39
• /
• 2019

In this paper, we propose a parametric model for analyzing the motion information obtained from the acceleration sensors to measure the activity of the human body. The motion of the upper body and the lower body does not occur at the same time, and the motion analysis method using a single motion sensor involves a lot of errors. In this study, the 3-axis accelerometer is attached to the arms and legs, the body's activity data are measured, the momentum of the arms and legs are calculated for each channel, and the linear predictive coefficient is obtained for each channel. The periodicity of the upper body and the lower body is determined by analyzing the correlation between the channels. The linear predictive coefficient and the periodic value are used as data to measure the type of exercise and the amount of exercise. In the proposed method, we measured four types of movements such as walking, stair climbing, slow hill climbing, and fast hill descending. In order to verify the usefulness of the parameters, the recognition results are presented using the linear predictive coefficient and the periodic value for each motion as the neural network input.

• Journal of the Korean Society of Manufacturing Process Engineers
• /
• v.18 no.9
• /
• pp.29-35
• /
• 2019

The structure of the machinery industry due to the 4th industrial revolution is changing from precision and durability to intelligent and smart machinery through sensing and interconnection(IoT). There is a growing need for research on prognostics and health management(PHM) that can prevent abnormalities in processing machines and accurately predict and diagnose conditions. PHM is a technology that monitors the condition of a mechanical system, diagnoses signs of failure, and predicts the remaining life of the object. In this study, the vibration generated during machining is measured and a classification algorithm for normal and fault signals is developed. Arbitrary fault signal is collected by changing the conditions of un stable supply cutting oil and fixing jig. The signal processing is performed to apply the measured signal to the learning model. The sampling rate is changed for high speed operation and performed machine learning using raw signal without FFT. The fault classification algorithm for 1D convolution neural network composed of 2 convolution layers is developed.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2018.05a
• /
• pp.45-45
• /
• 2018

오염물의 혼합거동을 해석하기 위해 물리기반 모델을 이용하는 경우 모델을 구축하고 운용하는데 많은 시간과 재정이 소요되며 현장검증을 통한 검증이 반드시 필요하다. 하지만 데이터 기반 모델의 경우 축적된 데이터만으로도 예측을 수행할 수 있으며 물리기반모델에 비해 결정해야할 입력인자가 적어 모델운용이 용이하다는 장점이 있다. 다양한 데이터 모델 중 인공신경망(ANN) 모델은 데이터가 가지는 불확실성 및 비정상성, 복잡한 상호관련성에 효과적으로 대응할 수 있는 모델로 수자원 및 환경 분야에서 자주 사용되고 있다. 본 연구에서는 인공신경망 모델을 이용하여 지천유입이 있는 대하천의 수질인자 (pH, 전기전도도, DO, chl-a)를 예측하였다. 다른 데이터기반 모델과 같이 인공신경망 모델 또한 수집된 데이터 질에 크게 영향을 받으며, 내부 입력인자의 선택이 모델의 예측 결과에 큰 영향을 미친다. 이러한 인공신경망 모델의 특성을 바탕으로 예측모형의 정확도를 향상하기 위해서는 크게 데이터 처리부분과 모델구축 부분에서의 접근이 필요하다. 본 연구에서는 데이터 처리 과정에서 연구대상지점의 각각의 수질인자가 가지는 분포 특성을 유지하기 위해 층화표츨추출법을 이용하여 데이터를 구성하였다. 모델의 구축 과정에서는 초기가중치 값의 영향을 줄이기 위해 앙상블기법을 사용하였으며, 좀 더 견고하고 정확한 결과를 예측하기 위해 탄력적 역전파알고리즘을 추가하였다. 추가적으로 합류 후 본류의 미 계측지역 수질 예측 정확도 향상을 위해 본류의 수질인자뿐만 아니라 지류의 수질인자를 입력자료로 사용하여 모의를 수행하였다. 또한 동일 구간에서 수행한 현장추적자실험 자료를 이용하여 수질인자의 분포특성을 비교, 검증하였다. 개발된 모델을 이용하여 낙동강과 금호강 합류부 하류의 수질인자를 예측한 결과 지류의 수질인자를 입력자료로 추가한 경우 예측의 정확도가 증가하였으며, 현장실험 자료를 통해 밝혀진 오염물의 거동현상을 인공신경망 모델로도 동일하게 재현하는 것으로 나타났다. 본 연구에서 제안한 인공신경모델을 이용한다면 물리기반 수치모델을 대체하여 지천으로 유입된 오염물의 거동을 정확하고 효율적으로 파악할 수 있을 것이다.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2019.05a
• /
• pp.363-363
• /
• 2019

활성화 함수(activation function)는 기계학습(machine learning)의 학습과정에 비선형성을 도입하여 심층적인 학습을 용이하게 하고 예측의 정확도를 높이는 중요한 요소 중 하나이다(Roy et al., 2019). 일반적으로 기계학습에서 사용되고 있는 활성화 함수의 종류에는 계단 함수(step function), 시그모이드 함수(sigmoid 함수), 쌍곡 탄젠트 함수(hyperbolic tangent function), ReLU 함수(Rectified Linear Unit function) 등이 있으며, 예측의 정확도 향상을 위하여 다양한 형태의 활성화 함수가 제시되고 있다. 본 연구에서는 기계학습을 통하여 수위예측 시 정확도 향상을 위하여 Hybrid 활성화 함수를 제안하였다. 연구대상지는 조수간만의 영향을 받는 한강을 대상으로 선정하였으며, 2009년 ~ 2018년까지 10년간의 수문자료를 활용하였다. 수위예측 알고리즘은 Python 내 Tensorflow의 RNN (Recurrent Neural Networks) 모델을 이용하였으며, 강수량, 수위, 조위, 댐 방류량, 하천 유량의 수문자료를 학습시켜 3시간 및 6시간 후의 수위를 예측하였다. 예측정확도 향상을 위하여 입력 데이터는 정규화(Normalization)를 시켰으며, 민감도 분석을 통하여 신경망모델의 은닉층 개수, 학습률의 최적 값을 도출하였다. Hybrid 활성화 함수는 쌍곡 탄젠트 함수와 ReLU 함수를 혼합한 형태로 각각의 가중치($w_1,w_2,w_1+w_2=1$)를 변경하여 정확도를 평가하였다. 그 결과 가중치의 비($w_1/w_2$)에 따라서 예측 결과의 RMSE(Roote Mean Square Error)가 최소가 되고 NSE (Nash-Sutcliffe model Efficiency coefficient)가 최대가 되는 지점과 Peak 수위의 예측정확도가 최대가 되는 지점을 확인할 수 있었다. 본 연구는 현재 Data modeling을 통한 수위예측의 정확도 향상을 위해 기초가 되는 연구이나, 향후 다양한 형태의 활성화 함수를 제안하여 정확도를 향상시킨다면 예측 결과를 통하여 침수예보에 대한 의사결정이 가능할 것으로 기대된다.

• Journal of the Korean Society of Propulsion Engineers
• /
• v.17 no.2
• /
• pp.71-83
• /
• 2013

Recently, the advanced condition monitoring methods such as the model-based method and the artificial intelligent method have been applied to maximize the availability as well as to minimize the maintenance cost of the aircraft gas turbines. Among them the non-linear GPA(Gas Path Analysis) method and the GA(Genetic Algorithms) have lots of advantages to diagnose the engines compared to other advanced condition monitoring methods such as the linear GPA, fuzzy logic and neural networks. Therefore this work applies both the non-linear GPA and the GA to diagnose AE3007 turbofan engine for an aircraft, and in case of having sensor noise and bias it is confirmed that the GA is better than the GPA through the comparison of two methods.

• Safety and Health at Work
• /
• v.2 no.4
• /
• pp.336-347
• /
• 2011

Objectives: In Korea, an average of 258 workers claim compensation for their noise-induced hearing loss (NIHL) on an annual basis. Indeed, hearing disorder ranks first in the number of diagnoses made by occupational medical check-ups. Against this backdrop, this study analyzed the impact of 19 types of noise-generating machines and equipment on the sound pressure levels in workplaces and NIHL occurrence based on a 2009 national survey on work environments. Methods: Through this analysis, a series of statistical models were built to determine posterior probabilities for each worksite with an aim to present risk ratings for noise levels at work. Results: It was found that air compressors and grinding machines came in first and second, respectively in the number of installed noise-generating machines and equipment. However, there was no direct relationship between workplace noise and NIHL among workers since noise-control equipment and protective gear had been in place. By building a logistic regression model and neural network, statistical models were set to identify the influence of the noise-generating machines and equipment on workplace noise levels and NIHL occurrence. Conclusion: This study offered NIHL prevention measures which are fit for the worksites in each risk grade.

• Journal of Korea Multimedia Society
• /
• v.22 no.10
• /
• pp.1168-1177
• /
• 2019

In order to diagnose and prevent Alzheimer's Disease (AD), it is becoming increasingly important to develop a CAD(Computer-aided Diagnosis) system for AD diagnosis, which provides effective treatment for patients by analyzing 3D MRI images. It is essential to apply powerful deep learning algorithms in order to automatically classify stages of Alzheimer's Disease and to develop a Alzheimer's Disease support diagnosis system that has the function of detecting hippocampus and CSF(Cerebrospinal fluid) which are important biomarkers in diagnosis of Alzheimer's Disease. In this paper, for AD diagnosis, we classify a given MRI data into three categories of AD, mild cognitive impairment, and normal control according by applying 3D brain MRI image to the Faster R-CNN model and detect hippocampus and CSF in MRI image. To do this, we use the 2D MRI slice images extracted from the 3D MRI data of the Faster R-CNN, and perform the widely used majority voting algorithm on the resulting bounding box labels for classification. To verify the proposed method, we used the public ADNI data set, which is the standard brain MRI database. Experimental results show that the proposed method achieves impressive classification performance compared with other state-of-the-art methods.

• The Journal of the Acoustical Society of Korea
• /
• v.40 no.5
• /
• pp.466-472
• /
• 2021

Speaker Change Detection (SCD) refers to finding the moment when the main speaker changes from one person to the next in a speech conversation. In speaker change detection, difficulties arise due to overlapping speakers, inaccuracy in the information labeling, and data imbalance. To solve these problems, TIMIT corpus widely used in speech recognition have been concatenated artificially to obtain a sufficient amount of training data, and the detection of changing speaker has performed after identifying overlapping speakers. In this paper, we propose an speaker change detection system that considers the speaker overlapping. We evaluated and verified the performance using various approaches. As a result, a detection system similar to the X-Vector structure was proposed to remove the speaker overlapping region, while the Bi-LSTM method was selected to model the speaker change system. The experimental results show a relative performance improvement of 4.6 % and 13.8 % respectively, compared to the baseline system. Additionally, we determined that a robust speaker change detection system can be built by conducting related studies based on the experimental results, taking into consideration text and speaker information.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.20 no.4
• /
• pp.43-49
• /
• 2019

The wired and wireless Internet is a useful window to easily acquire various types of media data. On the other hand, the public can easily get the media data including the object to which the personal information is exposed, which is a social problem. In this paper, we propose a method to robustly detect a target object that has exposed personal information using a learning algorithm and effectively block the detected target object area. In the proposed method, only the target object containing the personal information is detected using a neural network-based learning algorithm. Then, a grid-like mosaic is created and overlapped on the target object area detected in the previous step, thereby effectively blocking the object area containing the personal information. Experimental results show that the proposed algorithm robustly detects the object area in which personal information is exposed and effectively blocks the detected area through mosaic processing. The object blocking method presented in this paper is expected to be useful in many applications related to computer vision.