• Korean Journal of Remote Sensing
• /
• v.33 no.4
• /
• pp.423-436
• /
• 2017

This paper presents a deep learning-based road segmentation framework from very high-resolution orthophotos. The proposed method uses Deep Convolutional Autoencoders for end-to-end mapping of orthophotos to road segmentations. In addition, a set of post-processing steps were applied to make the model outputs GIS-ready data that could be useful for various applications. The optimization of the model's parameters is explained which was conducted via grid search method. The model was trained and implemented in Keras, a high-level deep learning framework run on top of Tensorflow. The results show that the proposed model with the best-obtained hyperparameters could segment road objects from orthophotos at an average accuracy of 88.5%. The results of optimization revealed that the best optimization algorithm and activation function for the studied task are Stochastic Gradient Descent (SGD) and Exponential Linear Unit (ELU), respectively. In addition, the best numbers of convolutional filters were found to be 8 for the first and second layers and 128 for the third and fourth layers of the proposed network architecture. Moreover, the analysis on the time complexity of the model showed that the model could be trained in 4 hours and 50 minutes on 1024 high-resolution images of size $106{\times}106pixels$, and segment road objects from similar size and resolution images in around 14 minutes. The results show that the deep learning models such as Convolutional Autoencoders could be a best alternative to traditional machine learning models for road segmentation from aerial photographs.

• Smart Structures and Systems
• /
• v.25 no.5
• /
• pp.605-617
• /
• 2020

The existence of damages in structures causes changes in the physical properties by reducing the modal parameters. In this paper, we develop a two-stages approach based on normalized Modal Strain Energy Damage Indicator (nMSEDI) for quick applications to predict the location of damage. A two-dimensional IsoGeometric Analysis (2D-IGA), Machine Learning Algorithm (MLA) and optimization techniques are combined to create a new tool. In the first stage, we introduce a modified damage identification technique based on frequencies using nMSEDI to locate the potential of damaged elements. In the second stage, after eliminating the healthy elements, the damage index values from nMSEDI are considered as input in the damage quantification algorithm. The hybrid of Teaching-Learning-Based Optimization (TLBO) with Artificial Neural Network (ANN) and Particle Swarm Optimization (PSO) are used along with nMSEDI. The objective of TLBO is to estimate the parameters of PSO-ANN to find a good training based on actual damage and estimated damage. The IGA model is updated using experimental results based on stiffness and mass matrix using the difference between calculated and measured frequencies as objective function. The feasibility and efficiency of nMSEDI-PSO-ANN after finding the best parameters by TLBO are demonstrated through the comparison with nMSEDI-IGA for different scenarios. The result of the analyses indicates that the proposed approach can be used to determine correctly the severity of damage in beam structures.

• Transactions of the Korean Society of Mechanical Engineers A
• /
• v.38 no.5
• /
• pp.489-495
• /
• 2014

A micro-genetic algorithm (MGA) is one of the improved forms of a genetic algorithm. It is used to reduce the number of iterations and the computing resources required by using small populations. The efficiency of MGAs has been proved through many problems, especially problems with 3-5 design variables. This study proposes an optimization algorithm based on the sequential design of experiments (SDOE) and an MGA. In a previous study, the authors used the SDOE technique to reduce trial-and-error in the conventional approximate optimization method by using the statistical design of experiments (DOE) and response surface method (RSM) systematically. The proposed algorithm has been applied to various mathematical examples and a structural problem.

• Transactions of the Korean Society of Mechanical Engineers A
• /
• v.35 no.10
• /
• pp.1323-1328
• /
• 2011

In the current design process for the lower arm used in automobile parts, an optimal solution of its various design variables should be found through exploration of the design space approximated using the response surface model formulated with a first- or second-order polynomial equation. In this study, a multi-level computational DOE (design of experiment) was carried out to explore the design space showing nonlinear behavior, in terms of factors such as the total weight and applied stress of the lower arm, where a fractional-factorial orthogonal array based on the artificial neural network model was introduced. In addition, the tolerance robustness of the optimal solution was estimated using a tolerance optimization with six sigma constraints, taking into account the tolerances occurring in the design variables.

• Transactions of the Korean Society of Automotive Engineers
• /
• v.15 no.1
• /
• pp.8-15
• /
• 2007

Freeway corridors consist of urban freeways and parallel arterials that drivers can use alternatively. Ramp metering in freeways and signal control in arterials are contemporary traffic control methods that have been developed and applied in order to improve traffic conditions of freeway corridors. However, most of the existing studies have focused on either optimal ramp metering in freeways, or progression signal strategies between arterial intersections. There have been no traffic control systems in Korea that integrates the freeway ramp metering and arterial signal control. The effective control strategies for freeway operations may cause negative effects on arterial traffic. On the other hand, traffic congestion and bottleneck phenomenon of arterials due to the increasing peak-hour travel demand and ineffective signal operation may generate an accessibility problem to freeway ramps. Thus, the main function of the freeway which is the through-traffic process has not been successful. The purpose of this study is to develop an integrated control model that connects freeway ramp metering systems and signal control systems in arterial intersections. And Optimization of integrated control model which consists of ramp metering and signal control is another purpose. The design of experiment, neural network, and simulated annealing are used for optimization.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.23 no.3
• /
• pp.326-332
• /
• 2019

Recently, the deep learning technology has gained lots of attention which leads to its application to various fields. Especially, there are recent attempts to overcome the limit of wireless communications systems through the use of the deep learning. In this paper, we have verified the performance of deep learning based transmit power control scheme. Unlike previous transmit power control schemes where the optimal transmit power is derived by solving the optimization problem explicitly, in the deep learning based transmit power control, the general solver for the optimization problem is derived through the deep neural network (DNN). Especially, by using the spectral efficiency as the loss function of DNN, the training can be performed without needing labels. Through simulation based on Tensorflow, we confirm that the transmit power control based on deep learning can achieve the optimal performance while reducing the computational complexity by 1/200.

• Journal of Biosystems Engineering
• /
• v.29 no.4
• /
• pp.341-346
• /
• 2004

Recent livestock people concern not only increase of production, but also superior quality of animal-breeding environment. So far, the optimization of the breeding and air environment has been focused on the production increase. In the very near future, the optimization will be emphasized on the environment for the animal welfare and health. Especially, cattle farming demands the precision livestock farming and special attention has to be given to the management of feeding, animal health and fertility. The management of individual animal is the first step for precision livestock farming and animal welfare, and recognizing each individual is important for that. Though electronic identification of a cattle such as RFID(Radio Frequency Identification) has many advantages, RFID implementations practically involve several problems such as the reading speed and distance. In that sense, computer vision might be more effective than RFID for the identification of an individual animal. The researches on the identification of cattle via image processing were mostly performed with the cows having black-white patterns of the Holstein. But, the native Korean and Japanese cattle do not have any definite pattern on the body. The purpose of this research is to identify the Japanese black cattle that does not have a body pattern using computer vision technology and neural network algorithm. Twelve heads of Japanese black cattle have been tested to verify the proposed scheme. The values of input parameters were specified and then computed using the face images of cattle. The images of cattle faces were trained using associate neural network algorithm, and the algorithm was verified by the face images that were transformed using brightness, distortion, and noise factors. As a result, there was difference due to transform ratio of the brightness, distortion, and noise. And, the proposed algorithm could identify 100% in the range from -3 to +3 degrees of the brightness, from -2 to +4 degrees of the distortion, and from 0% to 60% of the noise transformed images. It is concluded that our system can not be applied in real time recognition of the moving cows, but can be used for the cattle being at a standstill.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.30 no.6
• /
• pp.1291-1300
• /
• 2020

Recently, a deep learning-based non-profiling side-channel analysis was proposed. The deep learning-based non-profiling analysis is a technique that trains a neural network model for all guessed keys and then finds the correct secret key through the difference in the training metrics. As the performance of non-profiling analysis varies greatly depending on the neural network training model design, a correct model design criterion is required. This paper describes the two types of loss functions and eight labeling methods used in the training model design. It predicts the analysis performance of each labeling method in terms of non-profiling analysis and power consumption model. Considering the characteristics of non-profiling analysis and the HW (Hamming Weight) power consumption model is assumed, we predict that the learning model applying the HW label without One-hot encoding and the Correlation Optimization (CO) loss will have the best analysis performance. And we performed actual analysis on three data sets that are Subbytes operation part of AES-128 1 round. We verified our prediction by non-profiling analyzing two data sets with a total 16 of MLP-based model, which we describe.

• Journal of the Korean Society of Radiology
• /
• v.18 no.5
• /
• pp.531-540
• /
• 2024

Recently, research and development on automatic diagnosis solutions in the medical imaging field using deep learning are actively underway. In this study, we sought to find a fast and accurate classification deep learning modeling for classification of pneumonia in chest images using Inception V3, a deep learning model based on a convolutional artificial neural network. For this reason, after applying the optimization algorithms AdaGrad, RMS Prop, and Adam to deep learning modeling, deep learning modeling was implemented by selectively applying learning rates of 0.01 and 0.001, and then the performance of chest X-ray image pneumonia classification was compared and evaluated. As a result of the study, in verification modeling that can evaluate the performance of the classification model and the learning state of the artificial neural network, it was found that the performance of deep learning modeling for classification of the presence or absence of pneumonia in chest X-ray images was the best when applying Adam as the optimization algorithm with a learning rate of 0.001. I was able to. And in the case of Adam, which is mainly applied as an optimization algorithm when designing deep learning modeling, it showed excellent performance and excellent metric results when selectively applying learning rates of 0.01 and 0.001. In the metric evaluation of test modeling, AdaGrad, which applied a learning rate of 0.1, showed the best results. Based on these results, when designing deep learning modeling for binary-based medical image classification, in order to expect quick and accurate performance, a learning rate of 0.01 is preferentially applied when applying Adam as an optimization algorithm, and a learning rate of 0.01 is preferentially applied when applying AdaGrad. I recommend doing this. In addition, it is expected that the results of this study will be presented as basic data during similar research in the future, and it is expected to be used as useful data in the health and bio industries for the purpose of automatic diagnosis of medical images using deep learning.

• Journal of the Computational Structural Engineering Institute of Korea
• /
• v.35 no.6
• /
• pp.367-374
• /
• 2022

This papter presents the use of the automatic differential method based on the backpropagation method to obtain the design sensitivity and its application to topology optimization considering the stress constraints. Solving topology optimization problems with stress constraints is difficult owing to singularities, the local nature of stress constraints, and nonlinearity with respect to design variables. To solve the singularity problem, the stress relaxation technique is used, and p-norm for stress constraints is applied instead of local stresses for global stress measures. To overcome the nonlinearity of the design variables in stress constraint problems, it is important to analytically obtain the exact design sensitivity. In conventional topology optimization, design sensitivity is obtained efficiently and accurately using the adjoint variable method; however, obtaining the design sensitivity analytically and additionally solving the adjoint equation is difficult. To address this problem, the design sensitivity is obtained using a backpropagation technique that is used to determine optimal weights and biases in the artificial neural network, and it is applied to the topology optimization with the stress constraints. The backpropagation technique is used in automatic differentiation and can simplify the calculation of the design sensitivity for the objectives or constraint functions without complicated analytical derivations. In addition, the backpropagation process is more computationally efficient than solving adjoint equations in sensitivity calculations.