• The Journal of the Acoustical Society of Korea
• /
• v.39 no.5
• /
• pp.497-504
• /
• 2020

Constructing speaker embeddings in speaker verification is an important issue. In general, a self-attention mechanism has been applied for speaker embedding encoding. Previous studies focused on training the self-attention in a high-level layer, such as the last pooling layer. In this case, the effect of low-level layers is not well represented in the speaker embedding encoding. In this study, we propose Masked Cross Self-Attentive Encoding (MCSAE) using ResNet. It focuses on training the features of both high-level and low-level layers. Based on multi-layer aggregation, the output features of each residual layer are used for the MCSAE. In the MCSAE, the interdependence of each input features is trained by cross self-attention module. A random masking regularization module is also applied to prevent overfitting problem. The MCSAE enhances the weight of frames representing the speaker information. Then, the output features are concatenated and encoded in the speaker embedding. Therefore, a more informative speaker embedding is encoded by using the MCSAE. The experimental results showed an equal error rate of 2.63 % using the VoxCeleb1 evaluation dataset. It improved performance compared with the previous self-attentive encoding and state-of-the-art methods.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.21 no.12
• /
• pp.320-330
• /
• 2020

Fire is difficult to achieve good performance in image detection using deep learning because of its high irregularity. In particular, there is little data on fire detection in underground utility facilities, which have poor light conditions and many objects similar to fire. These make fire detection challenging and cause low performance of deep learning models. Therefore, this study proposed a fire detection model using deep learning and estimated the performance of the model. The proposed model was designed using a combination of a basic convolutional neural network, Inception block of GoogleNet, and Skip connection of ResNet to optimize the deep learning model for fire detection under underground utility facilities. In addition, a training technique for the model was proposed. To examine the effectiveness of the method, the trained model was applied to fire images, which included fire and non-fire (which can be misunderstood as a fire) objects under the underground facilities or similar conditions, and results were analyzed. Metrics, such as precision and recall from deep learning models of other studies, were compared with those of the proposed model to estimate the model performance qualitatively. The results showed that the proposed model has high precision and recall for fire detection under low light intensity and both low erroneous and missing detection capabilities for things similar to fire.

• Journal of agriculture & life science
• /
• v.45 no.3
• /
• pp.1-7
• /
• 2011

This study was carried out to provide the basic information for effective preservation and management of forest community of Nari basin in Ulleung Island. Forest community in Nari basin was classified into Fagus engleriana community, Sorbus amurensis community, Pinus densiflora community, Celtis jessoensis community and Alnus maximowiczii community. As the result of DCCA ordination analysis, sea level among environmental factors had high correlation with community distribution. Fagus engleriana community and Sorbus amurensis community correlated highly with aspect, Na content, and C/N ratio. There was a high correlation between Celtis jessoensis community and the content of Ca and K. Alnus maximowiczii community was distributed in site where CEC content is high. Pinus densiflora community was distributed in site where the content of Ca and CEC is high.

• Journal of Information Processing Systems
• /
• v.1 no.1 s.1
• /
• pp.70-74
• /
• 2005

Sensor networks have emerged as an interesting and important research area in the last few years. These networks require that time be synchronized more precisely than in traditional Internet applications. In this paper, we compared and analyzed the performance of the RBS and TDP mechanisms in the view of the number of generated messages and the synchronization accuracy. The reason that we chose be RBS ad the TDP mechanism to be compared is because the RES is an innovative method to achieve the high accurate synchronization. And TDP is a new method taking over the NTP method which has been used widely in the Internet. We simulated the performance of two methods assuming the IEEE 802.11 CSMA/CA MAC. As for the number of nodes in the sensor networks, two situations of 25 (for the small size network) and 100 (for the large size network) nodes are used. In the aspect of the number of messages generated for the synchronization, TDP is far better than RBS. But, the synchronization accuracy of RBS is far higher than that of TDP. We cm conclude that in a small size sensor networks requiring very high accuracy, such as an application of very high speed objects tracking in a confined space, the RBS is more proper than TDP even though the RBS may generate more traffic than TDP. But, in a wide range sensor networks with a large number of nodes, TDP is more realistic though the accuracy is somewhat worse than RBS because RBS may make so many synchronization messages, and then consume more energies at each node. So, two mechanisms may be used selectively according to the required environments, without saying that the one method is always better than the other.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.21 no.5
• /
• pp.197-202
• /
• 2021

Biometric technology using finger veins is receiving a lot of attention due to its high security, convenience and accuracy. And the recent development of deep learning technology has improved the processing speed and accuracy for authentication. However, the training data is a subset of real data not in a certain order or method and the results are not constant. so the amount of data and the complexity of the artificial neural network must be considered. In this paper, the deep learning model of Inception-Resnet-v2 was used to improve the high accuracy of the finger vein recognizer and the performance of the authentication system, We compared and analyzed the performance of the deep learning model of DenseNet-201. The simulations used data from MMCBNU_6000 of Jeonbuk National University and finger vein images taken directly. There is no preprocessing for the image in the finger vein authentication system, and the results are checked through EER.

• The Journal of The Korea Institute of Intelligent Transport Systems
• /
• v.18 no.1
• /
• pp.56-73
• /
• 2019

Probe-based systems have been gaining popularity in advanced traveler information systems. However, the high possibility of providing inaccurate travel-time information due to the inherent time-lag phenomenon is still an important issue to be resolved. To mitigate the time-lag problem, different prediction techniques have been applied, but the techniques are generally regarded as less effective for travel times with high variability. For this reason, current 5-min aggregated data have been commonly used for real-time travel-time provision on highways with high travel-time fluctuation. However, the 5-min aggregation interval itself can further increase the time-lags in the real-time travel-time information equivalent to 5 minutes. In this study, a new scheme that uses both individual probe and 5-min aggregated travel times is suggested to provide reliable real-time travel-time information. The scheme utilizes individual probe data under congested conditions and 5-min aggregated data under uncongested conditions, respectively. As a result of an evaluation with field data, the proposed scheme showed the best performance, with a maximum reduction in travel-time error of 18%.

• KOREAN JOURNAL OF CROP SCIENCE
• /
• v.44 no.4
• /
• pp.309-315
• /
• 1999

Rice seedling test was conducted to check the loging tolerance at ripening stage through evaluating the root characters. Thirteen Korean and foreign rice cultivars with direct seeding adaptable or high quality characteristics were grown in a cell pot and under submerged paddy. The root characters and pushing resistance of rice hill were determined at seedling and ripening stage, respectively. The diameter of crown root at the 7th and 8th leaf stages was thicker in lodging tolerance cultivars than those of others and showed significant-positive correlation with both pushing resistance and crown root diameter of mature plants. Also, the tensile strength of crown root at the 7th and 8th leaf stage showed highly positive correlation with the tensile strength of crown root of mature plants. The number of crown root at 7th leaf stage was significant-positively correlated with that of mature plant. The diameter of seminal root was not significantly correlated with the diameter of crown root throughout the whole growth stage. These results indicate that the diameter, tensile strength and number of crown root associated with root lodging tolerance can be detected with the seedling at about 7th or 8th leaf stage, and the seedling test using the cell pot is an useful and practical method to select lodging tolerant cultivars or lines of rice based on root characters, especially diameter of crown root.

• Proceedings of the Korean Society of Precision Engineering Conference
• /
• 2005.06a
• /
• pp.1980-1983
• /
• 2005

The characteristic of sheet metal process is the few loss of material during process, the short processing time and the excellent price and strength. The sheet metal process with above characteristic is common used in industrial field, but in order to analysis irregular field problems the reliable and economical analysis method is demanded. Finite element method is very effective method to simulate the forming processes with good prediction of the deformation behaviour. Among Finite element method, The static-implicit finite element method is applied effectively to analyze real-size auto-body panel stamping processes, which include the forming stage. In this paper, it was focussed on the drawability factors on auto-body panel stamping by AUTOFORM with using tool planing alloy to reduce law price as well as high precision from Design Optimization of ide. According to this study, the results of simulation will give engineers good information to access the Design Optimization of die.

• Journal of the Korean Society for Precision Engineering
• /
• v.23 no.3 s.180
• /
• pp.118-124
• /
• 2006

The characteristic of sheet metal process is the few loss of material during process, the short processing time and the excel lent price and strength. The sheet metal process with above characteristic is common used in industrial field, but in order to analysis irregular field problems the reliable and economical analysis method is demanded. Finite element method is very effective method to simulate the forming processes with good prediction of the deformation behavior. Among Finite element method, the static-implicit finite element method is applied effectively to analyze real-size auto-body panel stamping processes, which include the forming stage. In this paper, it was focused on the drawing ability factors on auto-body panel stamping by AUTOFORM with using tool planning alloy to reduce law price as well as high precision front Design Optimization of die. According to this study, the results of simulation will give engineers good information to access the Design Optimization of die.

• Journal of Korea Multimedia Society
• /
• v.23 no.6
• /
• pp.729-737
• /
• 2020

Screening with low-dose spiral computed tomography (LDCT) has been shown to reduce lung cancer mortality by about 20% when compared to standard chest radiography. One of the problems arising from screening programs is that large amounts of CT image data must be interpreted by radiologists. To solve this problem, automated detection of pulmonary nodules is necessary; however, this is a challenging task because of the high number of false positive results. Here we demonstrate detection of pulmonary nodules using six off-the-shelf convolutional neural network (CNN) models after modification of the input/output layers and end-to-end training based on publicly databases for comparative evaluation. We used the well-known CNN models, LeNet-5, VGG-16, GoogLeNet Inception V3, ResNet-152, DensNet-201, and NASNet. Most of the CNN models provided superior results to those of obtained using customized CNN models. It is more desirable to modify the proven off-the-shelf network model than to customize the network model to detect the pulmonary nodules.