• Journal of the Korea Convergence Society
• /
• v.12 no.5
• /
• pp.23-29
• /
• 2021

Recent deep learning-based natural language processing studies are conducting research to improve performance by training large amounts of data from various sources together. However, there is a possibility that the methodology of learning by combining data from various sources into one may prevent performance improvement. In the case of machine translation, data deviation occurs due to differences in translation(liberal, literal), style(colloquial, written, formal, etc.), domains, etc. Combining these corpora into one for learning can adversely affect performance. In this paper, we propose a new Corpus Weight Balance(CWB) method that considers the balance between parallel corpora in machine translation. As a result of the experiment, the model trained with balanced corpus showed better performance than the existing model. In addition, we propose an additional corpus construction process that enables coexistence with the human translation market, which can build high-quality parallel corpus even with a monolingual corpus.

• Journal of Convergence for Information Technology
• /
• v.11 no.11
• /
• pp.57-65
• /
• 2021

We propose a novel method to detect abnormal data of specific symptoms using deep learning in air pollution measurement system. Existing methods generally detect abnomal data by classifying data showing unusual patterns different from the existing time series data. However, these approaches have limitations in detecting specific symptoms. In this paper, we use DeepLab V3+ model mainly used for foreground segmentation of images, whose structure has been changed to handle one-dimensional data. Instead of images, the model receives time-series data from multiple sensors and can detect data showing specific symptoms. In addition, we improve model's performance by reducing the complexity of noisy form time series data by using 'piecewise aggregation approximation'. Through the experimental results, it can be confirmed that anomaly data detection can be performed successfully.

• 재활복지
• /
• v.21 no.3
• /
• pp.129-146
• /
• 2017

In an event of an emergency, such as fire in a building, visually impaired and blind people are prone to exposed to a level of danger that is greater than that of normal people, for they cannot be aware of it quickly. Current fire detection methods such as smoke detector is very slow and unreliable because it usually uses chemical sensor based technology to detect fire particles. But by using vision sensor instead, fire can be proven to be detected much faster as we show in our experiments. Previous studies have applied various image processing and machine learning techniques to detect fire, but they usually don't work very well because these techniques require hand-crafted features that do not generalize well to various scenarios. But with the help of recent advancement in the field of deep learning, this research can be conducted to help solve this problem by using deep learning-based object detector that can detect fire using images from security camera. Deep learning based approach can learn features automatically so they can usually generalize well to various scenes. In order to ensure maximum capacity, we applied the latest technologies in the field of computer vision such as YOLO detector in order to solve this task. Considering the trade-off between recall vs. complexity, we introduced two convolutional neural networks with slightly different model's complexity to detect fire at different recall rate. Both models can detect fire at 99% average precision, but one model has 76% recall at 30 FPS while another has 61% recall at 50 FPS. We also compare our model memory consumption with each other and show our models robustness by testing on various real-world scenarios.

• Journal of IKEEE
• /
• v.26 no.3
• /
• pp.488-493
• /
• 2022

In manufacturing sites, bearing fault in eletrically driven motors cause the entire system to shut down. Stopping the operation of this environment causes huge losses in time and money. The reason of this bearing defects can be various factors such as wear due to continuous contact of rotating elements, excessive load addition, and operating environment. In this paper, a motor driving environment is created which is similar to the domestic manufacturing sites. In addition, based on the established environment, we propose a dataset for bearing fault detection by collecting changes in vibration characteristics that vary depending on normal and defective conditions. The sensor used to collect the vibration characteristics is Microphone G.R.A.S. 40PH-10. We used various machine learning models to build a prototype bearing fault detection system trained on the proposed dataset. As the result, based on the deep neural network model, it shows high accuracy performance of 92.3% in the time domain and 98.3% in the frequency domain.

• ETRI Journal
• /
• v.44 no.2
• /
• pp.286-299
• /
• 2022

Human activity recognition in real time is a challenging task. Recently, a plethora of studies has been proposed using deep learning architectures. The implementation of these architectures requires the high computing power of the machine and a massive database. However, handcrafted features-based machine learning models need less computing power and very accurate where features are effectively extracted. In this study, we propose a handcrafted model based on three-dimensional sequential skeleton data. The human body skeleton movement over a frame is computed through joint positions in a frame. The joints of these skeletal frames are projected into two-dimensional space, forming a "movement polygon." These polygons are further transformed into a one-dimensional space by computing amplitudes at different angles from the centroid of polygons. The feature vector is formed by the sampling of these amplitudes at different angles. The performance of the algorithm is evaluated using a support vector machine on four public datasets: MSR Action3D, Berkeley MHAD, TST Fall Detection, and NTU-RGB+D, and the highest accuracies achieved on these datasets are 94.13%, 93.34%, 95.7%, and 86.8%, respectively. These accuracies are compared with similar state-of-the-art and show superior performance.

• Korean Journal of Biological Psychiatry
• /
• v.27 no.1
• /
• pp.18-26
• /
• 2020

Objectives The aim was to find effective vectorization and classification models to predict a psychiatric diagnosis from text-based medical records. Methods Electronic medical records (n = 494) of present illness were collected retrospectively in inpatient admission notes with three diagnoses of major depressive disorder, type 1 bipolar disorder, and schizophrenia. Data were split into 400 training data and 94 independent validation data. Data were vectorized by two different models such as term frequency-inverse document frequency (TF-IDF) and Doc2vec. Machine learning models for classification including stochastic gradient descent, logistic regression, support vector classification, and deep learning (DL) were applied to predict three psychiatric diagnoses. Five-fold cross-validation was used to find an effective model. Metrics such as accuracy, precision, recall, and F1-score were measured for comparison between the models. Results Five-fold cross-validation in training data showed DL model with Doc2vec was the most effective model to predict the diagnosis (accuracy = 0.87, F1-score = 0.87). However, these metrics have been reduced in independent test data set with final working DL models (accuracy = 0.79, F1-score = 0.79), while the model of logistic regression and support vector machine with Doc2vec showed slightly better performance (accuracy = 0.80, F1-score = 0.80) than the DL models with Doc2vec and others with TF-IDF. Conclusions The current results suggest that the vectorization may have more impact on the performance of classification than the machine learning model. However, data set had a number of limitations including small sample size, imbalance among the category, and its generalizability. With this regard, the need for research with multi-sites and large samples is suggested to improve the machine learning models.

• Journal of Korean Association for Spatial Structures
• /
• v.20 no.2
• /
• pp.51-58
• /
• 2020

Recently, deep learning that is the most popular and effective class of machine learning algorithms is widely applied to various industrial areas. A number of research on various topics about structural engineering was performed by using artificial neural networks, such as structural design optimization, vibration control and system identification etc. When nonlinear semi-active structural control devices are applied to building structure, a lot of computational effort is required to predict dynamic structural responses of finite element method (FEM) model for development of control algorithm. To solve this problem, an artificial neural network model was developed in this study. Among various deep learning algorithms, a recurrent neural network (RNN) was used to make the time history response prediction model. An RNN can retain state from one iteration to the next by using its own output as input for the next step. An eleven-story building structure with semi-active tuned mass damper (TMD) was used as an example structure. The semi-active TMD was composed of magnetorheological damper. Five historical earthquakes and five artificial ground motions were used as ground excitations for training of an RNN model. Another artificial ground motion that was not used for training was used for verification of the developed RNN model. Parametric studies on various hyper-parameters including number of hidden layers, sequence length, number of LSTM cells, etc. After appropriate training iteration of the RNN model with proper hyper-parameters, the RNN model for prediction of seismic responses of the building structure with semi-active TMD was developed. The developed RNN model can effectively provide very accurate seismic responses compared to the FEM model.

• Journal of Convergence for Information Technology
• /
• v.10 no.6
• /
• pp.41-48
• /
• 2020

The sensor nodes in underwater IoT networks have practical limitations in power supply. Thus, the reduction of power consumption is one of the most important issues in underwater environments. In this regard, AMC(Adaptive Modulation and Coding) techniques are used by using the relation between SNR and BER. However, according to our hands-on experience, we observed that the relation between SNR and BER is not that tight in underwater environments. Therefore, we propose a deep learning based MLP classification model to reflect multiple underwater channel parameters at the same time. It correctly predicts BER with a high accuracy of 85.2%. The proposed model can choose the best parameters to have the highest throughput. Simulation results show that the throughput can be enhanced by 4.4 times higher than the conventionally measured results.

• Smart Media Journal
• /
• v.9 no.3
• /
• pp.41-45
• /
• 2020

Recently, artificial intelligence (AI) and the Internet of things (IoT), which are represented by machine learning and deep learning among IT technologies related to the Fourth Industrial Revolution, are applied to our real life in various fields through various researches. In this paper, IoT and AI using object recognition technology are applied to classify clothing. For this purpose, the image dataset was taken using webcam and raspberry pi, and GoogLeNet, a convolutional neural network artificial intelligence network, was applied to transfer the photographed image data. The clothing image dataset was classified into two categories (shirtwaist, trousers): 900 clean images, 900 loss images, and total 1800 images. The classification measurement results showed that the accuracy of the clean clothing image was about 97.78%. In conclusion, the study confirmed the applicability of other objects using artificial intelligence networks on the Internet of Things based platform through the measurement results and the supplementation of more image data in the future.

• Journal of Broadcast Engineering
• /
• v.26 no.2
• /
• pp.175-183
• /
• 2021

The study of the fine grained classification of images continues to develop, but the study of object recognition for animals with polymorphic properties is proceeding slowly. Using only pet images corresponding to dogs and cats, this paper aims to compare methods using image processing and methods using deep learning among methods of classifying species of animals, which are fine grained classifications. In this paper, Grab-cut algorithm is used for object segmentation by method using image processing, and method using Fisher Vector for image encoding is proposed. Other methods used deep learning, which has achieved good results in various fields through machine learning, and among them, Convolutional Neural Network (CNN), which showed outstanding performance in image recognition, and Tensorflow, an open-source-based deep learning framework provided by Google. For each method proposed, 37 kinds of pet images, a total of 7,390 pages, were tested to verify and compare their effects.