• Journal of Biomedical Engineering Research
• /
• v.41 no.1
• /
• pp.48-54
• /
• 2020

In this paper, we deal with a liver fibrosis classification problem using ultrasound B-mode images. Commonly representative methods for classifying the stages of liver fibrosis include liver biopsy and diagnosis based on ultrasound images. The overall liver shape and the smoothness and roughness of speckle pattern represented in ultrasound images are used for determining the fibrosis stages. Although the ultrasound image based classification is used frequently as an alternative or complementary method of the invasive biopsy, it also has the limitations that liver fibrosis stage decision depends on the image quality and the doctor's experience. With the rapid development of deep learning algorithms, several studies using deep learning methods have been carried out for automated liver fibrosis classification and showed superior performance of high accuracy. The performance of those deep learning methods depends closely on the amount of datasets. We propose an enhanced U-net architecture to maximize the classification accuracy with limited small amount of image datasets. U-net is well known as a neural network for fast and precise segmentation of medical images. We design it newly for the purpose of classifying liver fibrosis stages. In order to assess the performance of the proposed architecture, numerical experiments are conducted on a total of 118 ultrasound B-mode images acquired from 78 patients with liver fibrosis symptoms of F0~F4 stages. The experimental results support that the performance of the proposed architecture is much better compared to the transfer learning using the pre-trained model of VGGNet.

• The Journal of Korea Institute of Information, Electronics, and Communication Technology
• /
• v.10 no.2
• /
• pp.147-153
• /
• 2017

Efficient texture representation is very important in computer vision fields. The performance of texture classification or/and segmentation can be improved based on efficient texture representation. Gabor filter is a representation method that has long history for texture representation based on multi-scale analysis. Gabor filter shows good performance in texture classification and segmentation but requires much processing time. In this paper, we propose new texture representation method that is also based on multi-scale analysis. The proposed representation can provide similar performance in texture classification but can reduce processing time against Gabor filter. Experimental results show good performance of our method.

• The Journal of the Acoustical Society of Korea
• /
• v.38 no.5
• /
• pp.568-573
• /
• 2019

In the diagnosis of pulmonary diseases, auscultation technique is simpler than the other methods, and lung sounds can be used for predicting the types of pulmonary diseases as well as identifying patients with pulmonary diseases. Therefore, in this paper, we identify patients with pulmonary diseases and classify lung sounds according to their sound characteristics using various convolutional neural networks, and compare the classification performance of each neural network method. First, lung sounds over affected areas of the chest with pulmonary diseases are collected by using a single-channel lung sound recording device, and spectral features are extracted from the collected sounds in time domain and applied to each neural network. As classification methods, we use general, parallel, and residual convolutional neural network, and compare lung sound classification performance of each neural network through experiments.

• Journal of the Korean Society for Library and Information Science
• /
• v.53 no.2
• /
• pp.179-199
• /
• 2019

In this paper, we propose optimal methodologies for classifying patent literature by examining various feature extraction methods, machine learning and deep learning models, and provide optimal performance through experiments. We compared the traditional BoW method and a distributed representation method (word embedding vector) as a feature extraction, and compared the morphological analysis and multi gram as the method of constructing the document collection. In addition, classification performance was verified using traditional machine learning model and deep learning model. Experimental results show that the best performance is achieved when we apply the deep learning model with distributed representation and morphological analysis based feature extraction. In Section, Class and Subclass classification experiments, We improved the performance by 5.71%, 18.84% and 21.53%, respectively, compared with traditional classification methods.

• Journal of Korean Library and Information Science Society
• /
• v.44 no.1
• /
• pp.27-47
• /
• 2013

This study investigated the classification performance of a kNN classifier using the feature selection methods based on document frequency(DF) and collection frequency(CF). The results of the experiments, which used HKIB-20000 data, were as follows. First, the feature selection methods that used high-frequency terms and removed low-frequency terms by the CF criterion achieved better classification performance than those using the DF criterion. Second, neither DF nor CF methods performed well when low-frequency terms were selected first in the feature selection process. Last, combining CF and DF criteria did not result in better classification performance than using the single feature selection criterion of DF or CF.

• Journal of Korea Multimedia Society
• /
• v.21 no.10
• /
• pp.1195-1202
• /
• 2018

Artificial neural network is widely used for its excellent performance and implementability. However, traditional neural network needs to learn the system from scratch, with the addition of new input data, the variation of the observation environment, or the change in the form of input/output data. To resolve such a problem, the technique of transfer learning has been proposed. Transfer learning constructs a newly developed target system partially updating existing system and hence provides much more efficient learning process. Until now, transfer learning is mainly studied in the field of image processing and is not yet widely employed in acoustic data processing. In this paper, focusing on the scalability of transfer learning, we apply the concept of transfer learning to the problem of guitar chord classification and evaluate its performance. For this purpose, we build a target system of convolutional neutral network (CNN) based 48 guitar chords classification system by applying the concept of transfer learning to a source system of CNN based 24 guitar chords classification system. We show that the system with transfer learning has performance similar to that of conventional system, but it requires only half the learning time.

• Journal of Satellite, Information and Communications
• /
• v.11 no.4
• /
• pp.52-57
• /
• 2016

The backscatter signals are very weak so they can be easily interfered by signal interferences and channels. In this paper, we propose a two-level classification algorithm for backscatter communications which chooses the idle frequency channel based on cognitive radio systems. The two-level classification algorithm provides an optimal idle frequency channel by obtaining informations about idle frequencies, fading of the channels, and the channels' usage state by primary users. Our simulation results show the improvement of BER and received power performance in backscatter communications by using the proposed algorithm, and the improvement of the algorithm's performance in backscatter communications.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.8 no.4
• /
• pp.1442-1462
• /
• 2014

Packet classification is a key technology of the Internet for routers to classify the arriving packets into different flows according to the predefined rulesets. Previous packet classification algorithms have mainly focused on search speed and memory usage, while overlooking update performance. In this paper, we propose PreCuts, which can drastically improve the update speed. According to the characteristics of IP field, we implement three heuristics to build a 3-layer decision tree. In the first layer, we group the rules with the same highest byte of source and destination IP addresses. For the second layer, we cluster the rules which share the same IP prefix length. Finally, we use the heuristic of information entropy-based bit partition to choose some specific bits of IP prefix to split the ruleset into subsets. The heuristics of PreCuts will not introduce rule duplication and incremental update will not reduce the time and space performance. Using ClassBench, it is shown that compared with BRPS and EffiCuts, the proposed algorithm not only improves the time and space performance, but also greatly increases the update speed.

• Proceedings of the Safety Management and Science Conference
• /
• 2006.11a
• /
• pp.293-297
• /
• 2006

Recently most universities are suffering from students leaving their majors. In order to make a countermeasure for reducing major separation rate, many universities are trying to find a proper solution. As a similar endeavor, this paper uses decision tree algorithm which is one of the data mining techniques which conduct grouping or prediction into several sub-groups from interested groups. This technique can analyze a feature of type on students leaving their majors. The dataset consists of 5,115 features through data selection from total data of 13,346 collected from a university in Kangwon-Do during seven years(2000.3.1 $\sim$ 2006.6.30). The main objective of this study is to evaluate performance of algorithms including CHAID, CART and C4.5 for classification of students leaving their majors with ROC Chart, Lift Chart and Gains Chart. Also, this study provides values about accuracy, sensitivity, specificity using classification table. According to the analysis result, CART showed the best performance for classification of students leaving their majors.

• Journal of the Korea Institute of Military Science and Technology
• /
• v.21 no.2
• /
• pp.158-165
• /
• 2018

Recently there is a great interest in the bi-static sonar. However, since the transmitter and the receiver operate on different platforms, it may be necessary to operate the system in a non-cooperative mode. In this situation, the detection and localization performance are limited. Therefore, it is necessary to classify the received pulse from the transmitter to overcome the performance limitation. In this paper, we proposed a robust automatic pulse classification method that can be applied to real systems. The proposed method eliminates the effects of noise and multipath propagation through post-processing and improves the pulse classification performance. We also verified the proposed method through the sea experimental data.