• Journal of KIISE:Software and Applications
• /
• v.26 no.11
• /
• pp.1261-1269
• /
• 1999

다층 퍼셉트론은 다양한 응용 분야에 성공적으로 적용되고 있는 대표적인 신경회로망 모델이다. 그러나 다층 퍼셉트론의 학습에서 나타나는 플라토에 기인한 느린 학습 속도와 지역 극소는 실제 응용문제에 적용함에 있어서 가장 큰 문제로 지적되어왔다. 이 문제를 해결하기 위해 여러 가지 다양한 학습알고리즘들이 개발되어 왔으나, 계산의 비효율성으로 인해 실제 문제에는 적용하기 힘든 예가 많은 등, 현재까지 만족할 만한 해결책은 제시되지 못하고 있다. 본 논문에서는 다층퍼셉트론의 베이시스 함수로 사용되는 시그모이드 함수를 보다 일반화된 형태로 정의하여 사용함으로써 학습에 있어서의 플라토를 완화하고, 지역극소에 빠지는 것을 줄이는 접근방법을 소개한다. 본 방법은 기존의 변형된 가중치 수정식을 사용한 학습 속도 향상의 방법들과는 다른 접근 방법을 택함으로써 기존의 방법들과 함께 사용하는 것이 가능하다는 특징을 갖고 있다. 제안하는 방법의 성능을 확인하기 위하여 간단한 패턴 인식 문제들에의 적용 실험 및 기존의 학습 속도 향상 방법을 함께 사용하여 시계열 예측 문제에 적용한 실험을 수행하였고, 그 결과로부터 제안안 방법의 효율성을 확인할 수 있었다. Abstract A multilayer perceptron is the most well-known neural network model which has been successfully applied to various fields of application. Its slow learning caused by plateau and local minima of gradient descent learning, however, have been pointed as the biggest problems in its practical use. To solve such a problem, a number of researches on learning algorithms have been conducted, but it can be said that none of satisfying solutions have been presented so far because the problems such as computational inefficiency have still been existed in these algorithms. In this paper, we propose a new learning approach to minimize the effect of plateau and reduce the possibility of getting trapped in local minima by generalizing the sigmoidal function which is used as the basis function of a multilayer perceptron. Adapting a new approach that differs from the conventional methods with revised updating equation, the proposed method can be used together with the existing methods to improve the learning performance. We conducted some experiments to test the proposed method on simple problems of pattern recognition and a problem of time series prediction, compared our results with the results of the existing methods, and confirmed that the proposed method is efficient enough to apply to the real problems.

• Journal of Bio-Environment Control
• /
• v.31 no.4
• /
• pp.270-278
• /
• 2022

This study was conducted using deep learning technology to classify for 'Mihwang' peach maturity with RGB images and fruit quality attributes during fruit development and maturation periods. The 730 images of peach were used in the training data set and validation data set at a ratio of 8:2. The remains of 170 images were used to test the deep learning models. In this study, among the fruit quality attributes, firmness, Hue value, and a^* value were adapted to the index with maturity classification, such as immature, mature, and over mature fruit. This study used the CNN (Convolutional Neural Networks) models for image classification; VGG16 and InceptionV3 of GoogLeNet. The performance results show 87.1% and 83.6% with Hue left value in VGG16 and InceptionV3, respectively. In contrast, the performance results show 72.2% and 76.9% with firmness in VGG16 and InceptionV3, respectively. The loss rate shows 54.3% and 62.1% with firmness in VGG16 and InceptionV3, respectively. It considers increasing for adapting a field utilization with firmness index in peach.

• Journal of the korean academy of Pediatric Dentistry
• /
• v.50 no.3
• /
• pp.263-276
• /
• 2023

The purpose of this study was to evaluate the performance of a model using You Only Look Once (YOLO) for object detection of proximal caries in periapical radiographs of children. A total of 2016 periapical radiographs in primary dentition were selected from the M6 database as a learning material group, of which 1143 were labeled as proximal caries by an experienced dentist using an annotation tool. After converting the annotations into a training dataset, YOLO was trained on the dataset using a single convolutional neural network (CNN) model. Accuracy, recall, specificity, precision, negative predictive value (NPV), F1-score, Precision-Recall curve, and AP (area under curve) were calculated for evaluation of the object detection model's performance in the 187 test datasets. The results showed that the CNN-based object detection model performed well in detecting proximal caries, with a diagnostic accuracy of 0.95, a recall of 0.94, a specificity of 0.97, a precision of 0.82, a NPV of 0.96, and an F1-score of 0.81. The AP was 0.83. This model could be a valuable tool for dentists in detecting carious lesions in periapical radiographs.

• Journal of the Korean Geosynthetics Society
• /
• v.23 no.2
• /
• pp.63-75
• /
• 2024

Drinking water is an essential element to ensure the basic human right to live, and the quality of clean water must always be ensured. However, domestic water facilities, which were installed intensively in the early 2000s, are deteriorating. The accidents such as discoloration of water such as chromaticity and turbidity as well as leakage of substances frequently occur. However, since it is virtually impossible to replace all water pipes, the detailed standards for maintenance of water pipe network facilities established in 2021 require water pipe cleaning. The swab pig method, one of the water pipe cleaning methods, is a method of physically removing substances in pipes and is evaluated as having the highest cleaning efficiency. However, Swab is highly likely to be damaged or deformed during the cleaning process, and may even be lost. Therefore, in this study, the material of the pig was changed to a material with high compressibility, and it was made as close as possible to the inner wall of the water pipe. And, to maximize cleaning efficiency, a rotation swab pig with a rotation blade was developed. In addition, high-strength wire and winding equipment were additionally developed to eliminate the possibility of loss and to determine the location of the pig. The inlet and outlet are connected with wires, and after verifying the performance of each detailed technology, the technology was applied on a test bed with a 30m section. As a result of the application, the performance of the technology was verified by measuring the process time and evaluating applicability.

• Information Systems Review
• /
• v.21 no.1
• /
• pp.23-38
• /
• 2019

The social network service (SNS) is one of the important marketing channels, so many companies actively exploit SNSs by posting SNS messages with appropriate content and style for their customers. In this paper, we focused on the psychological distances embedded in the SNS messages and developed a method to measure the psychological distance in SNS message by mixing a traditional content analysis, natural language processing (NLP), and machine learning. Through a traditional content analysis by human coding, the psychological distance was extracted from the SNS message, and these coding results were used for input data for NLP and machine learning. With NLP, word embedding was executed and Bag of Word was created. The Support Vector Machine, one of machine learning techniques was performed to train and test the psychological distance in SNS message. As a result, sensitivity and precision of SVM prediction were significantly low because of the extreme skewness of dataset. We improved the performance of SVM by balancing the ratio of data by upsampling technique and using data coded with the same value in first content analysis. All performance index was more than 70%, which showed that psychological distance can be measured well.

• Journal of Astronomy and Space Sciences
• /
• v.22 no.1
• /
• pp.1-12
• /
• 2005

The Korean VLBI Network (KVN) will open a new field of research in astronomy, geodesy and earth science using the newest three Elm radio telescopes. This will expand our ability to look at the Universe in the millimeter regime. Imaging capability of radio interferometry is highly dependent upon the antenna configuration, source size, declination and the shape of target. In this paper, imaging simulations are carried out with the KVN system configuration. Five test images were used which were a point source, multi-point sources, a uniform sphere with two different sizes compared to the synthesis beam of the KVN and a Very Large Array (VLA) image of Cygnus A. The declination for the full time simulation was set as +60 degrees and the observation time range was -6 to +6 hours around transit. Simulations have been done at 22GHz, one of the KVN observation frequency. All these simulations and data reductions have been run with the Astronomical Image Processing System (AIPS) software package. As the KVN array has a resolution of about 6 mas (milli arcsecond) at 220Hz, in case of model source being approximately the beam size or smaller, the ratio of peak intensity over RMS shows about 10000:1 and 5000:1. The other case in which model source is larger than the beam size, this ratio shows very low range of about 115:1 and 34:1. This is due to the lack of short baselines and the small number of antenna. We compare the coordinates of the model images with those of the cleaned images. The result shows mostly perfect correspondence except in the case of the 12mas uniform sphere. Therefore, the main astronomical targets for the KVN will be the compact sources and the KVN will have an excellent performance in the astrometry for these sources.

• Investigative Magnetic Resonance Imaging
• /
• v.9 no.1
• /
• pp.9-15
• /
• 2005

Purpose : This study attempted to investigate the effects of supply of highly concentrated $(30\%)$ oxygen on human ability of visuospatial cognition using time course data analysis of functional Magnetic Resonance Imaging (fMRI). Materials and Methods : To select an item set in the visuospatial performance test, two questionnaires with similar difficulty were developed through group testing. A group test was administered to 263 college students. Two types of questionnaire containing 20 questions were developed to measure the ability of visuospatial cognition. Eight college students (right-handed male, average age of 23.5 yrs) were examined for fMRI study. The experiment consisted of two runs of the visuospatial cognition testing, one with $21\%$ level of oxygen and the other with $30\%$ oxygen level. Each run consisted of 4 blocks, each containing control and visuospatial items. Functional brain images were taken from 37 MRI using the single-shot EPI method. Using the subtraction procedure, activated areas in the brain during visuospatial tasks were color-coded by t-score. To investigate the time course data in each activated area from brain images, 4 typical regions (cerebellum, occipital lobe, parietal lobe, and frontal lobe) were selected. Results : The average accuracy was $50.63{\pm}8.63$ and $62.50{\pm}9.64$ for $21\%\;and\;30\%$ oxygen respectively, and a statistically significant difference was found in the accuracy between the two types of oxygen (p<0.05). There were more activation areas observed at the cerebellum, occipital lobe, parietal lobe and frontal lobe with $30\%$ oxygen administration. The rate of increase in the cerebellum, occipital lobe and parietal lobe was $17\%$ and that of the frontal lobe, $50\%$. Especially, there were increase of intensity of BOLD signal at the parietal lobe with $30\%$ oxygen administration. The increase rate of the left parietal lobe was $1.4\%$ and that of the right parietal lobe, $1.7\%$. Conclusion : It is concluded that while performing visuospatial tasks, high concentrations of oxygen administration make oxygen administration sufficient, thus making neural network activate more, and the ability to perform visuospatial tasks increase.

• Journal of Korean Society of Transportation
• /
• v.20 no.4
• /
• pp.83-93
• /
• 2002

To improve the performance of estimation, the research that uses additional information addition to traffic count and target OD with additional survey cost have been studied. The purpose of this paper is to improve the performance of OD estimation by reducing the feasible solutions with cost-efficiently additional information addition to traffic counts and target OD. For this purpose, we Propose the OD estimation method with sample link use proportion as additional information. That is, we obtain the relationship between OD trip and link flow from sample link use proportion that is high reliable information with roadside survey, not from the traffic assignment of target OD. Therefore, this paper proposes OD estimation algorithm in which the conservation of link flow rule under the path-based non-equilibrium traffic assignment concept. Numerical result with test network shows that it is possible to improve the performance of OD estimation where the precision of additional data is low, since sample link use Proportion represented the information showing the relationship between OD trip and link flow. And this method shows the robust performance of estimation where traffic count or OD trip be changed, since this method did not largely affected by the error of target OD and the one of traffic count. In addition to, we also propose that we must set the level of data precision by considering the level of other information precision, because "precision problem between information" is generated when we use additional information like sample link use proportion etc. And we Propose that the method using traffic count as basic information must obtain the link flow to certain level in order to high the applicability of additional information. Finally, we propose that additional information on link have a optimal counting location problem. Expecially by Precision of information side it is possible that optimal survey location problem of sample link use proportion have a much impact on the performance of OD estimation rather than optimal counting location problem of link flow.

• The Journal of Engineering Geology
• /
• v.30 no.3
• /
• pp.237-251
• /
• 2020

This study examined structural analysis of supports in tunnel and displacement and underground stress of tunnel by measurement, in order to evaluate the performance of high-strength lattice girders developed as a substitute for H-profiles. According to the three-dimensional nonlinear structural analysis results of the tunnel support, the load and displacement relationship between the H-profiles and the high-strength lattice girders showed almost the same behavior, and the maximum load of the high-strength lattice girders were 1.0 to 1.2 times greater than the H-profiles. By the results of the three-dimensional tunnel cross-section analysis of the supports, the axial force was occurred largely in the lower left and right sides of the tunnel, and showed a similar trend to the field test values. In the results of the measurement of the roof settlement and rod extension, the final displacement of the steel arch rib (H-profile) and high-strength lattice girder section in tunnel was converged to a constant value without significant difference within the first management standard of 23.5 mm. According to the results of underground displacement measurement, the final change amount of the two support sections showed a slight displacement change, but converged to a constant value within the first management standard of 10 mm. By the results of measurement of shotcrete stress and steel arch rib stress, the final change amount of the two support sections showed a slight stress change, but converged to a constant value within the first management standard of 81.1 kg/㎠ and 54.2 tonf.

• The Journal of the Acoustical Society of Korea
• /
• v.43 no.1
• /
• pp.9-18
• /
• 2024

The importance of active sonar systems is emerging due to the quietness of underwater targets and the increase in ambient noise due to the increase in maritime traffic. However, the low signal-to-noise ratio of the echo signal due to multipath propagation of the signal, various clutter, ambient noise and reverberation makes it difficult to identify underwater targets using active sonar. Attempts have been made to apply data-based methods such as machine learning or deep learning to improve the performance of underwater target recognition systems, but it is difficult to collect enough data for training due to the nature of sonar datasets. Methods based on mathematical modeling have been mainly used to compensate for insufficient active sonar data. However, methodologies based on mathematical modeling have limitations in accurately simulating complex underwater phenomena. Therefore, in this paper, we propose a sonar signal synthesis method based on a deep neural network. In order to apply the neural network model to the field of sonar signal synthesis, the proposed method appropriately corrects the attention-based encoder and decoder to the sonar signal, which is the main module of the Tacotron model mainly used in the field of speech synthesis. It is possible to synthesize a signal more similar to the actual signal by training the proposed model using the dataset collected by arranging a simulated target in an actual marine environment. In order to verify the performance of the proposed method, Perceptual evaluation of audio quality test was conducted and within score difference -2.3 was shown compared to actual signal in a total of four different environments. These results prove that the active sonar signal generated by the proposed method approximates the actual signal.