• 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.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.15 no.4
• /
• pp.457-466
• /
• 2005

In this paper, we propose a wavelet neural network(WNN) based predictive control method for path tracking of mobile robots with multi-input and multi-output. In our control method, we use a WNN as a state predictor which combines the capability of artificial neural networks in learning processes and the capability of wavelet decomposition. A WNN predictor is tuned to minimize errors between the WNN outputs and the states of mobile robot using the gradient descent rule. And control signals, linear velocity and angular velocity, are calculated to minimize the predefined cost function using errors between the reference states and the predicted states. Through a computer simulation for the tracking performance according to varied track, we demonstrate the efficiency and the feasibility of our predictive control system.

• Nuclear Engineering and Technology
• /
• v.51 no.8
• /
• pp.1991-1997
• /
• 2019

In this manuscript, we present a method for the direct calculation of an ambient dose equivalent (H^* (10)) for the external gamma-ray exposure with an energy range of 40 keV to 2 MeV in an electronic personal dosimeter (EPD). The designed EPD consists of a 3 × 3 ㎟ PIN diode coupled to a 3 × 3 × 3 ㎣ CsI (Tl) scintillator block. The spectrum-to-dose conversion function (G(E)) for estimating H^* (10) was calculated by applying the gradient-descent method based on the Monte-Carlo simulation. The optimal parameters for the G(E) were found and this conversion of the H^* (10) from the gamma spectra was verified by using ²⁴¹Am, ¹³⁷Cs, ²²Na, ⁵⁴Mn, and ⁶⁰Co radioisotopes. Furthermore, gamma spectra and H^* (10) were obtained for an arbitrarily mixed multiple isotope case through Monte-Carlo simulation in order to expand the verification to more general cases. The H^* (10) based on the G(E) function for the gamma spectra was then compared with H^* (10) calculated by simulation. The relative difference of H^* (10) from various single-source spectra was in the range of ±2.89%, and the relative difference of H^* (10) for a multiple isotope case was in the range of ±5.56%.

• Journal of the Korea Society of Computer and Information
• /
• v.26 no.7
• /
• pp.1-7
• /
• 2021

In this paper, we propose a feature selection technique for multi-label classification. Many existing feature selection techniques have selected features by calculating the relation between features and labels such as a mutual information scale. However, since the mutual information measure requires a joint probability, it is difficult to calculate the joint probability from an actual premise feature set. Therefore, it has the disadvantage that only a few features can be calculated and only local optimization is possible. Away from this regional optimization problem, we propose a feature selection technique that constructs a low-rank space in the entire given feature space and selects features with sparsity. To this end, we designed a regression-based objective function using Nuclear norm, and proposed an algorithm of gradient descent method to solve the optimization problem of this objective function. Based on the results of multi-label classification experiments on four data and three multi-label classification performance, the proposed methodology showed better performance than the existing feature selection technique. In addition, it was showed by experimental results that the performance change is insensitive even to the parameter value change of the proposed objective function.

• Convergence Security Journal
• /
• v.23 no.5
• /
• pp.57-64
• /
• 2023

In this study, we analyzed how to steal the target model without training data. Input data is generated using the generative model, and a similar model is created by defining a loss function so that the predicted values of the target model and the similar model are close to each other. At this time, the target model has a process of learning so that the similar model is similar to it by gradient descent using the logit (logic) value of each class for the input data. The tensorflow machine learning library was used as an experimental environment, and CIFAR10 and SVHN were used as datasets. A similar model was created using the ResNet model as a target model. As a result of the experiment, it was found that the model stealing method generated a similar model with an accuracy of 86.18% for CIFAR10 and 96.02% for SVHN, producing similar predicted values to the target model. In addition, considerations on the model stealing method, military use, and limitations were also analyzed.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.37 no.2
• /
• pp.39-49
• /
• 2000

In this paper, we propose an optimizing feature extraction method for multiclass problems assuming normal distributions. Initially, We start with an arbitrary feature vector Assuming that the feature vector is used for classification, we compute the classification error Then we move the feature vector slightly in the direction so that classification error decreases most rapidly This can be done by taking gradient We propose two search methods, sequential search and global search In the sequential search, an additional feature vector is selected so that it provides the best accuracy along with the already chosen feature vectors In the global search, we are not constrained to use the chosen feature vectors Experimental results show that the proposed algorithm provides a favorable performance.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.17 no.3
• /
• pp.374-379
• /
• 2007

This paper focuses on the improvement of fault detection algorithm in FRTU(feeder remote terminal unit) on the feeder of distribution power system. FRTU is applied to fault detection schemes for phase fault and ground fault. Especially, cold load pickup and inrush restraint functions distinguish the fault current from the normal load current. FRTU shows FI(Fault Indicator) when the fault current is over pickup value or inrush current. STFT(Short Time Fourier Transform) analysis provides the frequency and time Information. FCM(Fuzzy C-Mean clustering) algorithm extracts characteristics of harmonics. The neural network system as a fault detector was trained to distinguish the inruih current from the fault status by a gradient descent method. In this paper, fault detection is improved by using FCM and neural network. The result data were measured in actual 22.9kV distribution power system.

• International Journal of Computer Science & Network Security
• /
• v.22 no.12
• /
• pp.185-196
• /
• 2022

The Internet of Things (IoT) has become more and more widespread in recent years, thus attackers are placing greater emphasis on IoT environments. The IoT connects a large number of smart devices via wired and wireless networks that incorporate sensors or actuators in order to produce and share meaningful information. Attackers employed IoT devices as bots to assault the target server; however, because of their resource limitations, these devices are easily infected with IoT malware. The Distributed Denial of Service (DDoS) is one of the many security problems that might arise in an IoT context. DDOS attempt involves flooding a target server with irrelevant requests in an effort to disrupt it fully or partially. This worst practice blocks the legitimate user requests from being processed. We explored an intelligent intrusion detection system (IIDS) using a particular sort of machine learning, such as Artificial Neural Networks, (ANN) in order to handle and mitigate this type of cyber-attacks. In this research paper Feed-Forward Neural Network (FNN) is tested for detecting the DDOS attacks using a modified version of the KDD Cup 99 dataset. The aim of this paper is to determine the performance of the most effective and efficient Back-propagation algorithms among several algorithms and check the potential capability of ANN- based network model as a classifier to counteract the cyber-attacks in IoT environments. We have found that except Gradient Descent with Momentum Algorithm, the success rate obtained by the other three optimized and effective Back- Propagation algorithms is above 99.00%. The experimental findings showed that the accuracy rate of the proposed method using ANN is satisfactory.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.20 no.1
• /
• pp.155-161
• /
• 2020

In general, optimization problems are difficult to solve simply. The reason is that the given problem is solved as soon as it is simple, but the more complex it is, the very large number of cases. This study is about the optimization of AI neural network. What we are dealing with here is the relief method for constructing AI network. The main topics deal with non-deterministic issues such as the stability and unstability of the overall network state, cost down and energy down. For this one, we discuss associative memory models, that is, a method in which local minimum memory information does not select fake information. The simulated annealing, this is a method of estimating the direction with the lowest possible value and combining it with the previous one to modify it to a lower value. And nonlinear planning problems, it is a method of checking and correcting the input / output by applying the appropriate gradient descent method to minimize the very large number of objective functions. This research suggests a useful approach to relief method as a theoretical approach to solving optimization problems. Therefore, this research will be a good proposal to apply efficiently when constructing a new AI neural network.

• Journal of the Korea Society of Computer and Information
• /
• v.29 no.2
• /
• pp.31-41
• /
• 2024

Deep learning models show excellent performance in tasks such as image classification and object detection in the field of computer vision, and are used in various ways in actual industrial sites. Recently, research on improving robustness has been actively conducted, along with pointing out that this deep learning model is vulnerable to hostile examples. A hostile example is an image in which small noise is added to induce misclassification, and can pose a significant threat when applying a deep learning model to a real environment. In this paper, we tried to confirm the robustness of the edge-learning classification model and the performance of the adversarial example detection model using it for adversarial examples of various algorithms. As a result of robustness experiments, the basic classification model showed about 17% accuracy for the FGSM algorithm, while the edge-learning models maintained accuracy in the 60-70% range, and the basic classification model showed accuracy in the 0-1% range for the PGD/DeepFool/CW algorithm, while the edge-learning models maintained accuracy in 80-90%. As a result of the adversarial example detection experiment, a high detection rate of 91-95% was confirmed for all algorithms of FGSM/PGD/DeepFool/CW. By presenting the possibility of defending against various hostile algorithms through this study, it is expected to improve the safety and reliability of deep learning models in various industries using computer vision.