• KIPS Transactions on Software and Data Engineering
• /
• v.9 no.7
• /
• pp.213-220
• /
• 2020

The convolutional neural network is composed of convolutional layers and fully connected layers. The nonlinear activation function is used in each layer of the convolutional layer and the fully connected layer. The activation function being used in a neural network is a function that simulates the method of transmitting information in a neuron that can transmit a signal and not send a signal if the input signal is above a certain criterion when transmitting a signal between neurons. The conventional activation function does not have a relationship with the loss function, so the process of finding the optimal solution is slow. In order to improve this, an agile activation function that generalizes the activation function is proposed. The agile activation function can improve the performance of the deep neural network in a way that selects the optimal agile parameter through the learning process using the primary differential coefficient of the loss function for the agile parameter in the backpropagation process. Through the MNIST classification problem, we have identified that agile activation functions have superior performance over conventional activation functions.

• KIPS Transactions on Software and Data Engineering
• /
• v.2 no.10
• /
• pp.705-712
• /
• 2013

Active learning aims to improve the performance of a classification model by repeating the process to select the most helpful unlabeled data and include it to the training set through labelling by expert. In this paper, we propose a method for active learning based on hierarchical agglomerative clustering using Ward's linkage. The proposed method is able to construct a training set actively so as to include at least one sample from each cluster and also to reflect the total data distribution by expanding the existing training set. While most of existing active learning methods assume that an initial training set is given, the proposed method is applicable in both cases when an initial training data is given or not given. Experimental results show the superiority of the proposed method.

• Korean Journal of Remote Sensing
• /
• v.6 no.2
• /
• pp.89-113
• /
• 1990

The purpose of this study is to develop means to apply GIS and remote sensing technology to the analysis of Korean urban open spaces. To achieve this objective, a framework of analysis of urban open spaces was developed, and then the framework was applied for the evaluation of the potential and suitability of open spaces of Ansan City, which is a new town developed to accomodate industries relocation from Seoul, Korea, mainly due to their pollution problems. The software used in this study are IDRISI, a grid-based GIS, and KMIPS, a remote sensing analysis system. Both packages are based on IBM PC/AT computers with Microsoft DOS. Landsat MSS and TM data were used for the land use classification, land use change detection, and analysis of transformed vegetation indices. The size of the geographic data base is 110 rows and 150 columns with the spatial resolution of 100m$\times$100m. The framework of analysis includes both quanititative and qualitative analysis of open spaces. The quantitative analysis includes size and distribution of open spaces, urban develpment of open spaces, and the degree of vegree of vegetation removal of the study area. The qualitative analysis includes evaluative criteria for primary productivity of land, park use potential, major visual resources, and urban environmental control. The findings of this study can be summarized as follows. First, the size of builtup areas increased 18.73km$^2$, while the size of forest land decreased 10.86km$^2$ during last ten years. Agricultural lands maintained its size, but shifted toward outside of the city into forest. Second, the potential of open spaces for park use is limited mainly due to their lack of accessibility and connectivity among open spaces, in spite of ample acreage and good site conditions. Third, major landscape elements and historic sites should be connected to the open space system of the city by new accesses and buffers.

• KIPS Transactions on Software and Data Engineering
• /
• v.10 no.10
• /
• pp.407-420
• /
• 2021

Deep neural networks are widely used to solve various problems. However, the deep neural network with a deep hidden layer frequently has a vanishing gradient or exploding gradient problem, which is a major obstacle to learning the deep neural network. In this paper, we propose a parametric activation function to alleviate the vanishing gradient problem that can be caused by nonlinear activation function. The proposed parametric activation function can be obtained by applying a parameter that can convert the scale and location of the activation function according to the characteristics of the input data, and the loss function can be minimized without limiting the derivative of the activation function through the backpropagation process. Through the XOR problem with 10 hidden layers and the MNIST classification problem with 8 hidden layers, the performance of the original nonlinear and parametric activation functions was compared, and it was confirmed that the proposed parametric activation function has superior performance in alleviating the vanishing gradient.

• Journal of the Korea Society of Computer and Information
• /
• v.26 no.12
• /
• pp.111-121
• /
• 2021

As the main carrier of information, digital image is becoming more and more important. However, with the popularity of image acquisition equipment and the rapid development of image editing software, in recent years, digital image counterfeiting incidents have emerged one after another, which not only reduces the credibility of images, but also brings great negative impacts to society and individuals. Image copy-paste tampering is one of the most common types of image tampering, which is easy to operate and effective, and is often used to change the semantic information of digital images. In this paper, a method to protect the authenticity and integrity of image content by studying the tamper detection method of image copy and paste was proposed. In view of the excellent learning and analysis ability of deep learning, two tamper detection methods based on deep learning were proposed, which use the traces left by image processing operations to distinguish the tampered area from the original area in the image. A series of experimental results verified the rationality of the theoretical basis, the accuracy of tampering detection, location and classification.

• Journal of the Korea Society of Computer and Information
• /
• v.26 no.7
• /
• pp.19-27
• /
• 2021

In this paper, we propose a method for detecting characters in images and detecting facial regions, which consists of two tasks. First, we separate two different characters to detect the face position of the characters in the frame. For fast detection, we use You Only Look Once (YOLO), which finds faces in the image in real time, to extract the location of the face and mark them as object detection boxes. Second, we present three image processing methods to detect accurate face area based on object detection boxes. Each method uses HSV values extracted from the region estimated by the detection figure to detect the face region of the characters, and changes the size and shape of the detection figure to compare the accuracy of each method. Each face detection method is compared and analyzed with comparative data and image processing data for reliability verification. As a result, we achieved the highest accuracy of 87% when using the split rectangular method among circular, rectangular, and split rectangular methods.

• KIPS Transactions on Software and Data Engineering
• /
• v.8 no.1
• /
• pp.19-26
• /
• 2019

In this paper, the classification performance of learning algorithms is compared for TAB digit recognition. The TAB digits that are segmented from TAB musical notes contain TAB lines and musical symbols. The labeling method and non-linear filter are designed and applied to extract fret digits only. The shift operation of the 4 directions is applied to generate more data. The selected models are Bayesian classifier, support vector machine, prototype based learning, multi-layer perceptron, and convolutional neural network. The result shows that the mean accuracy of the Bayesian classifier is about 85.0% while that of the others reaches more than 99.0%. In addition, the convolutional neural network outperforms the others in terms of generalization and the step of the data preprocessing.

• KIPS Transactions on Software and Data Engineering
• /
• v.7 no.11
• /
• pp.419-426
• /
• 2018

Due to the FMD(foot-and-mouth disease), the domestic animal husbandry and related industries suffer enormous damage every year. Although various academic researches related to FMD are ongoing, engineering studies on the social effects of FMD are very limited. In this study, we propose a systematic methodology to analyze emotional responses of regular citizens on FMD using text mining techniques. The proposed system first collects data related to FMD from the tweets posted on Twitter, and then performs a polarity classification process using a deep-learning technique. Second, keywords are extracted from the tweet using LDA, which is one of the typical techniques of topic modeling, and a keyword network is constructed from the extracted keywords. Finally, we analyze the various social effects of regular citizens on FMD through keyword network. As a case study, we performed the emotional analysis experiment of regular citizens about FMD from July 2010 to December 2011 in Korea.

• The Journal of Engineering Geology
• /
• v.28 no.4
• /
• pp.583-592
• /
• 2018

Recently, concern has arisen regarding the lowering of groundwater levels in the hinterland caused by the development of high-capacity radial collector wells in riverbank filtration areas. In this study, groundwater levels are estimated using Modflow software in relation to the water volume pumped by the radial collector well in Anseongcheon Stream. Using the water volume data, an artificial neural network (ANN) model is developed to determine the amount of water that can be withdrawn while minimizing the reduction of groundwater level. We estimate that increasing the pumping rate of the horizontal well HW-6, which is drilled parallel to the stream direction, is necessary to minimize the reduction of groundwater levels in wells OW-7 and OB-11. We also note that the number of input data and the classification of training and test data affect the results of the ANN model. This type of approach, which supplements ANN modeling with observed data, should contribute to the future groundwater management of hinterland areas.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.19 no.1
• /
• pp.1-8
• /
• 2019

Among the methods for extracting the most appropriate information from a large amount of log data, there is a method using inductive inference. In this paper, we use SVM (Support Vector Machine), which is an excellent classification method for inductive inference, in order to determine the ranking of intrusion logs in digital forensic analysis. For this purpose, the logs of the training log set are classified into intrusion logs and normal logs. The associated words are extracted from each classified set to generate a related word dictionary, and each log is expressed as a vector based on the generated dictionary. Next, the logs are learned using the SVM. We classify test logs into normal logs and intrusion logs by using the log set extracted through learning. Finally, the recommendation orders of intrusion logs are determined to recommend intrusion logs to the forensic analyst.