• Journal of Information Processing Systems
• /
• v.16 no.6
• /
• pp.1293-1308
• /
• 2020

In fine-tuning-based transfer learning, the size of the dataset may affect learning accuracy. When a dataset scale is small, fine-tuning-based transfer-learning methods use high computing costs, similar to a large-scale dataset. We propose a mid-level feature extractor that retrains only the mid-level convolutional layers, resulting in increased efficiency and reduced computing costs. This mid-level feature extractor is likely to provide an effective alternative in training a small-scale medical image dataset. The performance of the mid-level feature extractor is compared with the performance of low- and high-level feature extractors, as well as the fine-tuning method. First, the mid-level feature extractor takes a shorter time to converge than other methods do. Second, it shows good accuracy in validation loss evaluation. Third, it obtains an area under the ROC curve (AUC) of 0.87 in an untrained test dataset that is very different from the training dataset. Fourth, it extracts more clear feature maps about shape and part of the chest in the X-ray than fine-tuning method.

• The Journal of Society for e-Business Studies
• /
• v.12 no.3
• /
• pp.33-47
• /
• 2007

The requirements have been changed during development progresses, since it is impossible to define all of software requirements. These requirements change leads to mistakes because the developers cannot completely understand the software's structure and behavior, or they cannot discover all parts affected by a change. Requirement changes have to be managed and assessed to ensure that they are feasible, make economic sense and contribute to the business needs of the customer organization. We propose a feature-oriented requirements change management method to manage requirements change with value analysis and feature-oriented traceability links including intermediate catalysis using features. Our approach offers two contributions to the study of requirements change: (1) We define requirements change tree to make user requirements change request generalize by feature level. (2) We provide overall process such as change request normalization, change impact analysis, solution dealing with change request, change request implementation, change request evaluation. In addition, we especially present the results of a case study which is carried out in asset management portal system in details.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.21 no.4
• /
• pp.512-517
• /
• 2011

In this paper, we propose an Active Shape Model (ASM) and Lucas-Kanade (LK) optical flow-based feature extraction and analysis method for analyzing the emotional features from facial images. Considering the facial emotion feature regions are described by Facial Action Coding System, we construct the feature-related shape models based on the combination of landmarks and extract the LK optical flow vectors at each landmarks based on the centre pixels of motion vector window. The facial emotion features are modelled by the combination of the optical flow vectors and the emotional states of facial image can be estimated by the probabilistic estimation technique, such as Bayesian classifier. Also, we extract the optimal emotional features that are considered the high correlation between feature points and emotional states by using common spatial pattern (CSP) analysis in order to improvise the operational efficiency and accuracy of emotional feature extraction process.

• Smart Structures and Systems
• /
• v.12 no.3_4
• /
• pp.327-344
• /
• 2013

Beam string structure (BSS) is introduced as a new type of hybrid prestressed string structures. The composition and mechanics features of BSS are discussed. The main principles of wavelet packet transform (WPT), principal component analysis (PCA) and support vector machine (SVM) have been reviewed. WPT is applied to the structural response signals, and feature vectors are obtained by feature extraction and PCA. The feature vectors are used for training and classification as the inputs of the support vector machine. The method is used to a single one-way arched beam string structure for damage detection. The cable prestress loss and web members damage experiment for a beam string structure is carried through. Different prestressing forces are applied on the cable to simulate cable prestress loss, the prestressing forces are calculated by the frequencies which are solved by Fourier transform or wavelet transform under impulse excitation. Test results verify this method is accurate and convenient. The damage cases of web members on the beam are tested to validate the efficiency of the method presented in this study. Wavelet packet decomposition is applied to the structural response signals under ambient vibration, feature vectors are obtained by feature extraction method. The feature vectors are used for training and classification as the inputs of the support vector machine. The structural damage position and degree can be identified and classified, and the test result is highly accurate especially combined with principle component analysis.

• The Korean Journal of Applied Statistics
• /
• v.36 no.6
• /
• pp.501-514
• /
• 2023

When analyzing high dimensional data such as text data, if we input all the variables as explanatory variables, statistical learning procedures may suffer from over-fitting problems. Furthermore, computational efficiency can deteriorate with a large number of variables. Dimensionality reduction techniques such as feature selection or feature extraction are useful for dealing with these problems. The sparse principal component analysis (SPCA) is one of the regularized least squares methods which employs an elastic net-type objective function. The SPCA can be used to remove insignificant principal components and identify important variables from noisy observations. In this study, we propose a dimension reduction procedure for text data based on the SPCA. Applying the proposed procedure to real data, we find that the reduced feature set maintains sufficient information in text data while the size of the feature set is reduced by removing redundant variables. As a result, the proposed procedure can improve classification accuracy and computational efficiency, especially for some classifiers such as the k-nearest neighbors algorithm.

• Journal of Broadcast Engineering
• /
• v.27 no.3
• /
• pp.283-294
• /
• 2022

The feature map used in the network for deep learning generally has larger data than the image and a higher compression rate than the image compression rate is required to transmit the feature map. This paper proposes a method for transmitting a pyramid feature map with high compression rate, which is used in a network with an FPN structure that has robustness to object size in deep learning-based image processing. In order to efficiently compress the pyramid feature map, this paper proposes a structure that predicts a pyramid feature map of a level that is not transmitted with pyramid feature map of some levels that transmitted through the proposed prediction network to efficiently compress the pyramid feature map and restores compression damage through the proposed reconstruction network. Suggested mAP, the performance of object detection for the COCO data set 2017 Train images of the proposed method, showed a performance improvement of 31.25% in BD-rate compared to the result of compressing the feature map through VTM12.0 in the rate-precision graph, and compared to the method of performing compression through PCA and DeepCABAC, the BD-rate improved by 57.79%.

• Proceedings of the Korean Society for Noise and Vibration Engineering Conference
• /
• 2007.11a
• /
• pp.891-894
• /
• 2007

This study presents fault diagnosis of low speed bearing using support vector machine (SVM). The data used in the experiment was acquired using acoustic emission (AE) sensor and accelerometer. The aim of this study is to compare the performance of fault diagnosis based on AE signal and vibration signal with same load and speed. A low speed test rig was developed to simulate various defects with shaft speeds as low as 10 rpm under several loading conditions. In this study, component analysis was also performed to extract the feature and reduce the dimensionality of original data feature. Moreover, the classification for fault diagnosis was also conducted using original data feature without feature extraction. The result shows that extracted feature from AE sensor gave better performance in faults classification.

• Transactions of the Korean Society for Noise and Vibration Engineering
• /
• v.21 no.11
• /
• pp.1020-1028
• /
• 2011

As rotating machines play an important role in industrial applications such as aeronautical, naval and automotive industries, many researchers have developed various condition monitoring system and fault diagnosis system by applying artificial neural network. Since using obtained signals without preprocessing as inputs of neural network can decrease performance of fault classification, it is very important to extract significant features of captured signals and to apply suitable features into diagnosis system according to the kinds of obtained signals. Therefore, this paper proposes a neural-network-based fault diagnosis system using AR coefficients as feature vectors by LPC(linear predictive coding) and EIV(errors-in variables) analysis. We extracted feature vectors from sound, vibration and current faulty signals and evaluated the suitability of feature vectors depending on the classification results and training error rates by changing AR order and adding noise. From experimental results, we conclude that classification results using feature vectors by EIV analysis indicate more than 90 % stably for less than 10 orders and noise effect comparing to LPC.

• Journal of the Korea Society of Computer and Information
• /
• v.21 no.4
• /
• pp.63-71
• /
• 2016

This paper is about applying efficient data mining method which improves the score calculation and proper building performance of credit ranking score system. The main idea of this data mining technique is accomplishing such objectives by applying Correlation based Feature Selection which could also be used to verify the properness of existing rank scores quickly. This study selected 2047 manufacturing companies on KOSPI market during the period of 2009 to 2013, which have their own credit rank scores given by NICE information service agency. Regarding the relevant financial variables, total 80 variables were collected from KIS-Value and DART (Data Analysis, Retrieval and Transfer System). If correlation based feature selection could select more important variables, then required information and cost would be reduced significantly. Through analysis, this study show that the proposed correlation based feature selection method improves selection and classification process of credit rank system so that the accuracy and credibility would be increased while the cost for building system would be decreased.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.10 no.4
• /
• pp.687-692
• /
• 2006

In a conventional ICA(Independent Component Analysis) based face recognition method, PCA(Principal Component Analysis) first is used for feature extraction, ICA learning method then is applied for feature enhancement in the reduced dimension. It is not considered that a necessary component can be located in the discarded feature space. In the new ICA(NICA), learning extracts features using the magnitude of kurtosis (4-th order central moment or cumulant). But, the pure ICA method can not discard noise effectively. The synergy effect of PCA and ICA can be achieved if PCA is used for noise reduction filter. Namely, PCA does whitening and noise filtering. ICA performs feature extraction. Experiment results show the effectiveness of the new ICA method compared to the conventional ICA approach.