• The KIPS Transactions:PartB
• /
• v.14B no.4
• /
• pp.287-294
• /
• 2007

Many classification algorithms for real world data suffer from a data class imbalance problem. To solve this problem, various methods have been proposed such as altering the training balance and designing better sampling strategies. The previous methods are not satisfy in the distribution of the input data and the constraint. In this paper, we propose a focused sampling method which is more superior than previous methods. To solve the problem, we must select some useful data set from all training sets. To get useful data set, the proposed method devide the region according to scores which are computed based on the distribution of SOM over the input data. The scores are sorted in ascending order. They represent the distribution or the input data, which may in turn represent the characteristics or the whole data. A new training dataset is obtained by eliminating unuseful data which are located in the region between an upper bound and a lower bound. The proposed method gives a better or at least similar performance compare to classification accuracy of previous approaches. Besides, it also gives several benefits : ratio reduction of class imbalance; size reduction of training sets; prevention of over-fitting. The proposed method has been tested with kNN classifier. An experimental result in ecoli data set shows that this method achieves the precision up to 2.27 times than the other methods.

• Korean Journal of Ecology and Environment
• /
• v.33 no.3 s.91
• /
• pp.230-243
• /
• 2000

A large-scale community data were patterned by utilizing an unsupervised learning algorithm in artificial neural networks. Data for benthic macroinvertebrates in streams of South Korea reported in publications for 12 years from 1984 to 1995 were provided as inputs for training with the Kohonen network. Taxa included for the training were 5 phylum, 10 class, 26 order, 108 family and 571 species in 27 streams. Abundant groups were Diptera, Ephemeroptera, Trichoptera, Plecoptera, Coleoptera, Odonata, Oligochaeta, and Physidae. A wide spectrum of community compositions was observed: a few tolerant taxa were collected at polluted sites while a high species richness was observed at relatively clean sites. The trained mapping by the Kohonen network effectively showed patterns of communities from different river systems, followed by patterns of communities from different environmental disturbances. The training by the proposed artificial neural network could be an alternative for organizing community data in a large-scale ecological survey.

• Korean Journal of Agricultural and Forest Meteorology
• /
• v.16 no.2
• /
• pp.114-124
• /
• 2014

For classifying fire damaged areas and analyzing burn severity of two large-fire areas damaged over 100 ha in 2011, three methods were employed utilized supervised classification, unsupervised classification and Normalized Difference Vegetation Index (NDVI). In this paper, the post-fire imageries of SPOT were used to compute the Maximum Likelihood (MLC), Minimum Distance (MIN), ISODATA, K-means, NDVI and to evaluate large-scale patterns of burn severity from 1 m to 5 m spatial resolutions. The result of the accuracy verification on burn severity from satellite images showed that average overall accuracy was 88.38 % and the Kappa coefficient was 0.8147. To compare the accuracy between burn severity and field survey at Uljin and Youngduk, two large fire sites were selected as study areas, and forty-four sampling plots were assigned in each study area for field survey. The burn severities of the study areas were estimated by analyzing burn severity (BS) classes from SPOT images taken one month after the occurrence of the fire. The applicability of composite burn index (CBI) was validated with a correlation analysis between field survey data and burn severity classified by SPOT5, and by their confusion matrix. The result showed that correlation between field survey data and BS by SPOT5 were closely correlated in both Uljin (r = -0.544 and p<0.01) and Youngduk (r = -0.616 and p<0.01). Thus, this result supported that the proposed burn severity analysis is an adequate method to measure burn severity of large fire areas in Korea.

• Journal of Korea Water Resources Association
• /
• v.34 no.4
• /
• pp.303-316
• /
• 2001

In this study, Radial Basis Function(RBF) Neural Networks Model, a kind of Hybrid Neural Networks was applied to hydrological forecasting in a small watershed. RBF Neural Networks Model has four kinds of parameters in it and consists of unsupervised and supervised training patterns. And Gaussian Kernel Function(GKF) was used among many kinds of Radial Basis Functions(RBFs). K-Means clustering algorithm was applied to optimize centers and widths which ate the parameters of GKF. The parameters of RBF Neural Networks Model such as centers, widths weights and biases were determined by the training procedures of RBF Neural Networks Model. And, with these parameters the validation procedures of RBF Neural Networks Model were carried out. RBF Neural Networks Model was applied to Wi-Stream basin which is one of the IHP Representative basins in South Korea. 10 rainfall events were selected for training and validation of RBF Neural Networks Model. The results of RBF Neural Networks Model were compared with those of Elman Neural Networks(ENN) Model. ENN Model is composed of One Step Secant BackPropagation(OSSBP) and Resilient BackPropagation(RBP) algorithms. RBF Neural Networks shows better results than ENN Model. RBF Neural Networks Model spent less time for the training of model and can be easily used by the hydrologists with little background knowledge of RBF Neural Networks Model.

• Journal of KIISE:Computing Practices and Letters
• /
• v.7 no.6
• /
• pp.549-558
• /
• 2001

It is important to reuse existing motion capture data for reduction of the animation producing cost as well as efficiency of producing process. Because its motion curve has no control point, however, it is difficult to modify the captured data interactively. The motion transition is a useful method to reuse the existing motion data. It generates a seamless intermediate motion with two short motion sequences. In this paper, Uniform Posture Map (UPM) algorithm is proposed to perform the motion transition. Since the UPM is organized through quantization of various postures with an unsupervised learning algorithm, it places the output neurons with similar posture in adjacent position. Using this property, an intermediate posture of two active postures is generated; the generating posture is used as a key-frame to make an interpolating motion. The UPM algorithm needs much less computational cost, in comparison with other motion transition algorithms. It provides a control parameter; an animator could control the motion simply by adjusting the parameter. These merits of the UPM make an animator to produce the animation interactively. The UPM algorithm prevents from generating an unreal posture in learning phase. It not only makes more realistic motion curves, but also contributes to making more natural motions. The motion transition algorithm proposed in this paper could be applied to the various fields such as real time 3D games, virtual reality applications, web 3D applications, and etc.

• Journal of the Korea Society of Computer and Information
• /
• v.15 no.2
• /
• pp.65-74
• /
• 2010

We propose a system that automatically recognizes the tense or relaxed condition of scrolling-shooting game subject that plays. Existing study compares the changed values of source of stimulation to the player by suggesting the source, and thus involves limitation in automatic classification. This study applies SOM of unsupervised learning for automatic classification and recognition of player's condition change. Application of SOM for automatic recognition of tense and relaxed condition is composed of two steps. First, ECG measurement and analysis, is to extract characteristic vector through HRV analysis by measuring ECG after having the player play the game. Secondly, SOM learning and recognition, is to classify and recognize the tense and relaxed conditions of player through SOM learning of the input vectors of heart beat signals that the characteristic extracted. Experiment results are divided into three groups. The first is HRV frequency change and the second the SOM learning results of heart beat signal. The third is the analysis of match rate to identify SOM learning performance. As a result of matching the LF/HF ratio of HRV frequency analysis to the distance of winner neuron of SOM based on 1.5, a match rate of 72% performance in average was shown.

• Journal of the Korean Association of Geographic Information Studies
• /
• v.16 no.2
• /
• pp.88-100
• /
• 2013

Urban green space is one of most important aspects of urban infrastructure for improving the quality of life of city dwellers as it reduces the heat island effect and is used for recreation and relaxation. However, no systematic management of urban green space has been introduced in Korea as past practices focused on efficient development. A way to calculate the amount of green space needed to complement an urban area must be developed to preserve urban green space and to determine 'regulations determining the total amount of greenery'. In recent years, various studies have quantified urban green space and infrastructure using remotely sensed data. However, it is difficult to detect a myriad small green spaces in a city effectively when considering the spatial resolution of the data used in existing research. In this paper, we quantified small urban green spaces using CASI-1500 hyperspectral imagery. We calculated MCARI, a vegetation index for hyperspectral imagery, to evaluate the greenness of small green spaces. In addition, we applied image-classification methods, including the ISODATA algorithm and Spectral Angle Mapper, to detect small green spaces using supervised and unsupervised classifications. This could be used to categorize land-cover into four classes: unclassified, impervious, suspected green, and vegetation green.

• Journal of Cadastre & Land InformatiX
• /
• v.50 no.1
• /
• pp.201-214
• /
• 2020

This study aimed to early detect damaged trees by pine wilt disease using the vegetation indices of UAV images. The location data of 193 pine wilt disease trees were constructed through field surveys and vegetation index analyses of NDVI, GNDVI, NDRE and SAVI were performed using multi-spectral UAV images at the same time. K-Means algorithm was adopted to classify damaged trees and confusion matrix was used to compare and analyze the classification accuracy. The results of the study are summarized as follows. First, the overall accuracy of the classification was analyzed in order of NDVI (88.04%, Kappa coefficient 0.76) > GNDVI (86.01%, Kappa coefficient 0.72) > NDRE (77.35%, Kappa coefficient 0.55) > SAVI (76.84%, Kappa coefficient 0.54) and showed the highest accuracy of NDVI. Second, K-Means unsupervised classification method using NDVI or GNDVI is possible to some extent to find out the damaged trees. In particular, this technique is to help early detection of damaged trees due to its intensive operation, low user intervention and relatively simple analysis process. In the future, it is expected that the utilization of time series images or the application of deep learning techniques will increase the accuracy of classification.

• Journal of Intelligence and Information Systems
• /
• v.16 no.2
• /
• pp.33-54
• /
• 2010

The World Wide Web is transitioning from being a mere collection of documents that contain useful information toward providing a collection of services that perform useful tasks. The emerging Web service technology has been envisioned as the next technological wave and is expected to play an important role in this recent transformation of the Web. By providing interoperable interface standards for application-to-application communication, Web services can be combined with component based software development to promote application interaction both within and across enterprises. To make Web services for service-oriented computing operational, it is important that Web service repositories not only be well-structured but also provide efficient tools for developers to find reusable Web service components that meet their needs. As the potential of Web services for service-oriented computing is being widely recognized, the demand for effective Web service discovery mechanisms is concomitantly growing. A number of public Web service repositories have been proposed, but the Web service taxonomy generation has not been satisfactorily addressed. Unfortunately, most existing Web service taxonomies are either too rudimentary to be useful or too hard to be maintained. In this paper, we propose a Web service taxonomy generation framework that combines an artificial neural network based clustering techniques with descriptive label generating and leverages the semantics of the XML-based service specification in WSDL documents. We believe that this is one of the first attempts at applying data mining techniques in the Web service discovery domain. We have developed a prototype system based on the proposed framework using an unsupervised artificial neural network and empirically evaluated the proposed approach and tool using real Web service descriptions drawn from operational Web service repositories. We report on some preliminary results demonstrating the efficacy of the proposed approach.

• Journal of the Korean association of regional geographers
• /
• v.20 no.2
• /
• pp.206-216
• /
• 2014

In this study, forest types in Chungheongnam-do were analyzed using Landsat satellite images and digital forest type map as a means to estimate forest carbon stocks. NDVI and Tasseled Cap, ISODATA, and supervised classification among others were used to analyze the forest types. The forest carbon stocks of Chungcheongnam-do were estimated utilizing forest statistical data derived from the classified results. The results indicate that the analysis of forest types through supervised classification yielded the highest overall accuracy in analyzing forest types using satellite images. Coniferous forests(49.3%) accounted for the highest proportion in all the forest types of Chungcheongnam-do, followed by deciduous forests(28.0%) and mixed forests(22.7%). The results of a comparative analysis between forest carbon stocks estimates made using the modified digital forest type map and other estimation methods showed that the method using Tasseled Cap and unsupervised classification yielded the most similar forest carbon stock estimates. The most significant difference, though, was made when only the digital forest type map was used. It is expected that if carbon stocks are estimated by integrating satellite images and digital forest type maps in the future, more accurate results can be derived in estimating forest carbon stocks at a national level.