• Journal of the Korean Institute of Intelligent Systems
• /
• v.27 no.1
• /
• pp.22-28
• /
• 2017

The Convolutional Neural Network(CNN) can be divided into two stages: feature extraction and classification. The hyperparameters such as kernel size, number of channels, and stride in the feature extraction step affect the overall performance of CNN as well as determining the structure of CNN. In this paper, we propose a method to optimize the hyperparameter in CNN feature extraction stage using Parameter-Setting-Free Harmony Search (PSF-HS) algorithm. After setting the overall structure of CNN, hyperparameter was set as a variable and the hyperparameter was optimized by applying PSF-HS algorithm. The simulation was conducted using MATLAB, and CNN learned and tested using mnist data. We update the parameters for a total of 500 times, and it is confirmed that the structure with the highest accuracy among the CNN structures obtained by the proposed method classifies the mnist data with an accuracy of 99.28%.

• Journal of the Korean Society for Railway
• /
• v.18 no.6
• /
• pp.619-629
• /
• 2015

The train conflict resolution problem refers to an adjustment of the train operation schedule in order to minimize the delay propagation when a train conflict is predicted or when a train conflict occurs. Previous train studies of train conflict resolutions are limited in terms of the size of the problems to be solved due to exponential increases in the variables. In this paper, we propose a train conflict resolution model using a time-space network to solve the train conflict situation in the operational phase. The proposed model adjusts the size of the problem by giving only the dwell tolerance in the time-space network only for stops at the station after a train conflict. In addition, the proposed model can solve large problems using a path flow variable. The study presents a train delay propagation analysis and experimental results of train conflict resolution assessments as well using the proposed model.

• Journal of Digital Convergence
• /
• v.19 no.10
• /
• pp.253-263
• /
• 2021

Machine learning is the process of constructing a cost function using learning data used for learning and an artificial neural network to predict the data, and finding parameters that minimize the cost function. Parameters are changed by using the gradient-based method of the cost function. The more complex the digital signal and the more complex the problem to be learned, the more complex and deeper the structure of the artificial neural network. Such a complex and deep neural network structure can cause over-fitting problems. In order to avoid over-fitting, a weight decay regularization method of parameters is used. We additionally use the value of the cost function in this method. In this way, the accuracy of machine learning is improved, and the superiority is confirmed through numerical experiments. These results derive accurate values for a wide range of artificial intelligence data through machine learning.

• KSCE Journal of Civil and Environmental Engineering Research
• /
• v.39 no.1
• /
• pp.147-154
• /
• 2019

The finite element model updating can be defined as the problem of finding the parameters of the finite element model which gives the closest response to the actual response of the structure by measurement. In the previous researches, optimization based methods have been developed to minimize the error of the response of the actual structure and the analytical model. In this study, we propose an inverse eigenvalue problem that can directly obtain the parameters of the finite element model from the target mode information. Deep Neural Networks are constructed to solve the inverse eigenvalue problem quickly and accurately. As an application example of the developed method, the dynamic finite element model update of a suspension bridge is presented in which the deep neural network simulating the inverse eigenvalue function is utilized. The analysis results show that the proposed method can find the finite element model parameters corresponding to the target modes with very high accuracy.

• Journal of the Korea Convergence Society
• /
• v.10 no.7
• /
• pp.23-31
• /
• 2019

In this study, we collected data through online travel notes from January to December 2018 and analyzed the seasonal travel characteristics of individual visiting Chinese by utilizing social network analysis. The analysis showed that Seoul is a hub for Chinese travel to Korea and the main destinations for individual visiting Chinese are concentrated in Seoul, Busan, Jeju Island, Gyeongju and Gangneung, with wide differences in seasons. The research results can be used as basic data for the development of tourism courses for individual Chinese tourists to Korea, provision of tourism services and optimization of tourism facility layout. Future research can consider continuing to use network travel notes to study the tourist destination and the mode of transportation between tourist nodes, which can provide reference for the development of tourist market and the planning and design of tourist traffic.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.24 no.1
• /
• pp.50-56
• /
• 2020

With the rapid development of deep neural network theory and application research, the effect of face detection has been improved. However, due to the complexity of deep neural network calculation and the high complexity of the detection environment, how to detect face quickly and accurately becomes the main problem. This paper is based on the relatively simple model of the MTCNN model, using FDDB (Face Detection Dataset and Benchmark Homepage), LFW (Field Label Face) and FaceScrub public datasets as training samples. At the same time of sorting out and introducing MTCNN(Multi-Task Cascaded Convolutional Neural Network) model, it explores how to improve training speed and Increase performance at the same time. In this paper, the dynamic image pyramid technology is used to replace the traditional image pyramid technology to segment samples, and OHEM (the online hard example mine) function in MTCNN model is deleted in training, so as to improve the training speed.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.15 no.6
• /
• pp.987-994
• /
• 2020

In this paper, we studied the effect of kernel size of CNN layer on performance in acoustic noise attenuators. This system uses a deep learning algorithm using a neural network adaptive prediction filter instead of using the existing adaptive filter. Speech is estimated from a single input speech signal containing noise using a 100-neuron, 16-filter CNN filter and an error back propagation algorithm. This is to use the quasi-periodic property in the voiced sound section of the voice signal. In this study, a simulation program using Tensorflow and Keras libraries was written and a simulation was performed to verify the performance of the noise attenuator for the kernel size. As a result of the simulation, when the kernel size is about 16, the MSE and MAE values are the smallest, and when the size is smaller or larger than 16, the MSE and MAE values increase. It can be seen that in the case of an speech signal, the features can be best captured when the kernel size is about 16.

• Journal of Digital Convergence
• /
• v.19 no.6
• /
• pp.233-238
• /
• 2021

Recently, blockchain technology accumulates and stores all transactions. Therefore, in order to verify the contents of all transactions, the data itself is compressed, but the scalability is limited. In addition, since a separate verification algorithm is used for each type of transaction, the verification burden increases as the size of the transaction increases. Existing blockchain cannot participate in the network because it does not become a block sink by using a server with a low specification. Due to this problem, as the time passes, the data size of the blockchain network becomes larger and it becomes impossible to participate in the network except for users with abundant resources. Therefore, in this paper, we are improved transaction as studied the zero knowledge proof algorithm for general operation verification. In this system, the design of zero-knowledge circuit generator capable of general operation verification and optimization of verifier and prover were also conducted.

• International Journal of Computer Science & Network Security
• /
• v.21 no.1
• /
• pp.137-142
• /
• 2021

The Internet of Things (IoT) paradigm is at the forefront of present and future research activities. The huge amount of sensing data from IoT devices needing to be processed is increasing dramatically in volume, variety, and velocity. In response, cloud computing was involved in handling the challenges of collecting, storing, and processing jobs. The fog computing technology is a model that is used to support cloud computing by implementing pre-processing jobs close to the end-user for realizing low latency, less power consumption in the cloud side, and high scalability. However, it may be that some resources in fog computing networks are not suitable for some kind of jobs, or the number of requests increases outside capacity. So, it is more efficient to decrease sending jobs to the cloud. Hence some other fog resources are idle, and it is better to be federated rather than forwarding them to the cloud server. Obviously, this issue affects the performance of the fog environment when dealing with big data applications or applications that are sensitive to time processing. This research aims to build a fog topology job scheduling (FTJS) to schedule the incoming jobs which are generated from the IoT devices and discover all available fog nodes with their capabilities. Also, the fog topology job placement algorithm is introduced to deploy jobs into appropriate resources in the network effectively. Finally, by comparing our result with the state-of-art first come first serve (FCFS) scheduling technique, the overall execution time is reduced significantly by approximately 20%, the energy consumption in the cloud side is reduced by 18%.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.15 no.5
• /
• pp.1814-1828
• /
• 2021

Low-light image enhancement is a key technique to overcome the quality degradation of photos taken under scotopic vision illumination conditions. The degradation includes low brightness, low contrast, and outstanding noise, which would seriously affect the vision of the human eye recognition ability and subsequent image processing. In this paper, we propose an approach based on deep learning and Retinex theory to enhance the low-light image, which includes image decomposition, illumination prediction, image reconstruction, and image optimization. The first three parts can reconstruct the enhanced image that suffers from low-resolution. To reduce the noise of the enhanced image and improve the image quality, a super-resolution algorithm based on the Laplacian pyramid network is introduced to optimize the image. The Laplacian pyramid network can improve the resolution of the enhanced image through multiple feature extraction and deconvolution operations. Furthermore, a combination loss function is explored in the network training stage to improve the efficiency of the algorithm. Extensive experiments and comprehensive evaluations demonstrate the strength of the proposed method, the result is closer to the real-world scene in lightness, color, and details. Besides, experiments also demonstrate that the proposed method with the single low-light image can achieve the same effect as multi-exposure image fusion algorithm and no ghost is introduced.