• IEMEK Journal of Embedded Systems and Applications
• /
• v.15 no.2
• /
• pp.51-60
• /
• 2020

Optical character recognition (OCR) has been studied for decades because it is very useful in a variety of places. Nowadays, OCR's performance has improved significantly due to outstanding deep learning technology. Thus, there is an increasing demand for commercial-grade but affordable OCR systems. We have developed a low-cost, high-performance OCR system for the industry with the cheapest embedded developer kit that supports GPU acceleration. To achieve high accuracy for industrial use on limited computing resources, we chose a state-of-the-art text recognition algorithm that uses an end-to-end deep learning network as a baseline model. The model was then improved by replacing the feature extraction network with the best one suited to our conditions. Among the various candidate networks, EfficientNet-B3 has shown the best performance: excellent recognition accuracy with relatively low memory consumption. Besides, we have optimized the model written in TensorFlow's Python API using TensorFlow-TensorRT integration and TensorFlow's C++ API, respectively.

• Tunnel and Underground Space
• /
• v.27 no.5
• /
• pp.303-311
• /
• 2017

The correlation analysis between the results obtained from DFN flow model and equivalent continuum flow model were conducted on total of 72 DFN blocks having various fracture geometry and domain size. A strong linear relation seems to exist between the two approaches under condition that normalized relative error for continuum behavior (ER) is less than 0.2, and the results from both methods are found to almost identical. To explore the field applicability of equivalent continuum flow model in DFN media, a total of 48 numerical schemes related to inflow of underground circular openings were implemented under various DFN configurations. The equivalent continuum flow model in DFN media with a constant hydraulic aperture was evaluated as valid. However, as the anisotropy increases due to variation of the hydraulic aperture, the results are likely to be overestimated compare to the DFN flow model.

• Journal of the Korean Institute of Gas
• /
• v.15 no.1
• /
• pp.54-59
• /
• 2011

Generally, the flow hunting is observed in almost all of the orifice meters but the intensity of the flow hunting is different at each metering system. In order to investigate the relations between pipe diameter and the flow instability or the flow hunting in a real metering system, a one-dimensional pipeline network model was built and analyzed for the examination of flow characteristics and relations to the flow hunting, changing diameters of the meter and the pipes before and after the meter. Finally, the effects of pressuredifference variation and flow hunting following to the variations of the diameters of the meter and the pipes before and after the meter were investigated and the relations were examined as well.

• Korean Journal of Air-Conditioning and Refrigeration Engineering
• /
• v.20 no.6
• /
• pp.416-426
• /
• 2008

The boiler tube failure often experienced in the superheater of a utility boiler can seriously affect the economic and safe operation of the power plant. It has been known that this failure is mainly caused by the thermal load deviation in the superheater tube system, and deeply intensified by the non-uniform distribution of steam flow rates. The nonuniform steam flow is distinctively prominent at low power load rather than at full power load. In this paper, we analyze the steam flow distribution in the superheater tube system by using one dimensional flow network model. At 30% power load, the deviation of steam flow rate is predicted to be within 0.8% of the averaged flow rate. This deviation can be reduced to 0.1% and 0.07% by assuming two cases, that is, the removal of 13th tube at each tube rows and the installation of intermediate header, respectively. The assumed two cases would be effective for the uniform steam flow distribution across 85 superheater tube rows.

• Transactions of the Korean Society of Mechanical Engineers A
• /
• v.21 no.5
• /
• pp.828-834
• /
• 1997

The knowledge of material stress-strain behavior is an essential requirement for design and analysis of deformation processes. Empirical stress-strain relationship and constitutive equations describing material behavior during deformation are being widely used, despite suffering some drawbacks in terms of ease of development, accuracy and speed. In the present study, back-propagation neural networks are used to model and predict the flow stresses of a HSLA steel under conditions of constant strain, strain rate and temperature. The performance of the network model is comparedto those of statistical models on rate equations. Well-trained network model provides fast and accurate results, making it superior to statistical models.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.10 no.10
• /
• pp.4825-4847
• /
• 2016

In this paper, we focus on the per-flow throughput analysis of IEEE 802.11 multi-hop ad hoc networks. The importance of an accurate saturation throughput model lies in establishing the theoretical foundation for effective protocol performance improvements. We argue that the challenge in modeling the per-flow throughput in IEEE 802.11 multi-hop ad hoc networks lies in the analysis of the freezing process and probability of collisions. We first classify collisions occurring in the whole transmission process into instantaneous collisions and persistent collisions. Then we present a four-dimensional Markov chain model based on the notion of the fixed length channel slot to model the Binary Exponential Backoff (BEB) algorithm performed by a tagged node. We further adopt a continuous time Markov model to analyze the freezing process. Through an iterative way, we derive the per-flow throughput of the network. Finally, we validate the accuracy of our model by comparing the analytical results with that obtained by simulations.

• Journal of the Regional Association of Architectural Institute of Korea
• /
• v.20 no.6
• /
• pp.17-23
• /
• 2018

The purpose of this study is to suggest 100MPa class ultra high strength concrete mix design model applying neural network theory, in order to minimize an effort wasted by trials and errors method until now. Mix design model was applied to each of the 70 data using binary binder, ternary binder and quaternary binder. Then being repeatedly applied to back-propagation algorithm in neural network model, optimized connection weight was gained. The completed mix design model was proved, by analyzing and comparing to value predicted from mix design model and value measured from actual compressive strength test. According to the results of this study, more accurate value could be gained through the mix design model, if error rate decreases with the test condition and environment. Also if content of water and binder, slump flow, and air content of concrete apply to mix design model, more accurate and resonable mix design could be gained.

• Journal of the Institute of Electronics Engineers of Korea TC
• /
• v.41 no.4
• /
• pp.35-44
• /
• 2004

Recently announced TCP Vegas predicts the degree of congestion in the network and then control the congestion window size. Thus it shows better performance than TCP Reno. however, TCP vegas does not assume any network model, its congestion window control is very limited. Because or this limitation, TCP vegas still can not adapt to fast changing available bandwidth. In this paper, we introduce a new TCP algorithm which adapts to fast changing available bandwidth well. To devise such a TCP, we model the end to end network of TCP connection as a queueing system and finds congestion window size which can utilize the available bandwidth sufficiently but not make the network congested. The simulation results show that our algorithm adapts to the avaliable bandwidth faster than TCP vegas and as a results, when the available bandwidth is changing rapidly, our algorithm not only operates more stably than TCP Vegas, but also it shows higher thruput than TCP Vegas.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2018.05a
• /
• pp.258-258
• /
• 2018

In this study, a hydraulic flow model and an error correction model are combined to improve the flood simulation accuracy. First, the hydraulic flow model is calibrated by optimizing the Manning's roughness coefficient that considers spatial and temporal variability. Then, an error correction model were used to correct the systematic errors of the calibrated hydraulic model. The error correction model is developed using Artificial Neural Networks (ANNs) that can estimate the systematic simulation errors of the hydraulic model by considering some state variables as inputs. The input variables are selected using parital mutual information (PMI) technique. It was found that the calibrated hydraulic model can simulate flood water levels with good accuracy. Then, the accuracy of estimated flood levels is improved further by using the error correction model. The method proposed in this study can be used to the flood control and water resources management as it can provide accurate water level eatimation.

• Journal of the Economic Geographical Society of Korea
• /
• v.12 no.4
• /
• pp.540-556
• /
• 2009

The hub and spoke network is a critical network-based infrastructure that is widely applied in current transportation and telecommunications systems, including Internets, air transportation networks and highway systems. This main idea of hub location models is to construct a network system which achieves the economy of scale of flows. The main purpose of this study is to introduce new hub location problems that take into account network reliability. Two standard models based on assignment schemes are proposed, and a minimum threshold model is provided as an extension in terms of hub network design. The reliability and interaction potentials of 15 nodes in the U.S. are used to examine model behaviors. According to the type of models and reliability, hubs, and minimum threshold levels, relationships among the flow economy of scale, network costs, and network resiliency are analyzed.