• Wind and Structures
• /
• v.33 no.6
• /
• pp.423-435
• /
• 2021

Although existing algorithms can predict wind speed using historical observation data, for engineering feasibility, most use moment methods and probability density functions to estimate fitted parameters. However, extreme wind speed prediction accuracy for long-term return periods is not always dependent on how the optimized frequency distribution curves are obtained; long-term return periods emphasize general distribution effects rather than marginal distributions, which are closely related to potential extreme values. Moreover, there are different wind speed parent sample types; how to theoretically select the proper extreme value distribution is uncertain. The influence of different sampling time intervals has not been evaluated in the fitting process. To overcome these shortcomings, updated steps are introduced, involving parameter sensitivity analysis for different sampling time intervals. The extreme value prediction accuracy of unknown parent samples is also discussed. Probability analysis of mean wind is combined with estimation of the probability plot correlation coefficient and the maximum likelihood method; an iterative estimation algorithm is proposed. With the updated steps and comparison using a Monte Carlo simulation, a fitting policy suitable for different parent distributions is proposed; its feasibility is demonstrated in extreme wind speed evaluations at Longhua and Chuansha meteorological stations in Shanghai, China.

• Journal of the Korea Convergence Society
• /
• v.12 no.1
• /
• pp.49-55
• /
• 2021

Recently, with the development of database, it is possible to store a lot of data generated in finance, security, and networks. These data are being analyzed through classifiers based on machine learning. The main problem at this time is data imbalance. When we train imbalanced data, it may happen that classification accuracy is degraded due to over-fitting with majority class data. To overcome the problem of data imbalance, oversampling strategy that increases the quantity of data of minority class data is widely used. It requires to tuning process about suitable method and parameters for data distribution. To improve the process, In this study, we propose a strategy to explore and optimize oversampling combinations and ratio based on various methods such as synthetic minority oversampling technique and generative adversarial networks through genetic algorithms. After sampling credit card fraud detection which is a representative case of data imbalance, with the proposed strategy and single oversampling strategies, we compare the performance of trained classifiers with each data. As a result, a strategy that is optimized by exploring for ratio of each method with genetic algorithms was superior to previous strategies.

• Journal of Digital Contents Society
• /
• v.18 no.1
• /
• pp.159-165
• /
• 2017

In order to solve problems in automatic quality inspection of tablet capsules, computation-efficient image processing technique, appropriate threshold setting, edge detection and segmentation methods are required. And since existing automatic system for quality inspection of tablet capsules is of very high cost, it needs to be reduced through the realization of low-price hardware system. This study suggests a technique that uses low-cost camera module to obtain image and inspects dents on tablet capsules and sorting them by applying TLS curve fitting technique and edge-based image segmentation. In order to assess the performance, the major classifications algorithm of PCA, ICA and SVM are used to evaluate training time, test time and accuracy for capsule image area and curve fitting edge data sets.

• Journal of Korea Multimedia Society
• /
• v.21 no.2
• /
• pp.98-107
• /
• 2018

In this paper, we propose a method to classify the skeletal maturity with a small amount of hand wrist X-ray image using deep learning-based meta-learning. General deep-learning techniques require large amounts of data, but in many cases, these data sets are not available for practical application. Lack of learning data is usually solved through transfer learning using pre-trained models with large data sets. However, transfer learning performance may be degraded due to over fitting for unknown new task with small data, which results in poor generalization capability. In addition, medical images require high cost resources such as a professional manpower and mcuh time to obtain labeled data. Therefore, in this paper, we use meta-learning that can classify using only a small amount of new data by pre-trained models trained with various learning tasks. First, we train the meta-model by using a separate data set composed of various learning tasks. The network learns to classify the bone maturity using the bone maturity data composed of the radiographs of the wrist. Then, we compare the results of the classification using the conventional learning algorithm with the results of the meta learning by the same number of learning data sets.

• Proceedings of the KSRS Conference
• /
• 2007.10a
• /
• pp.162-165
• /
• 2007

Chlorophyll concentration is an important factor for physical oceanography as well as biological oceanography. For these necessity many oceanographic researchers have been investigated it for a long time. But investigation using vessel is very inefficient, on the other hands, ocean color remote sensing is a powerful means to get fine-scale (spatial and temporal scale) measurements of chlorophyll concentration. Geostationary Ocean Color Imager (GOCI), for ocean color sensor, loaded on COMS (Communication, Ocean and Meteorological Satellite), will be launched on late 2008 in Korea. According to the necessity of algorithm for GOCI, we developed chlorophyll algorithm for GOCI in this study. There are two types of chlorophyll algorithms. One is an empirical algorithm using band ratio, and the other one is a fluorescence-based algorithms. To develop GOCI chlorophyll algorithm empirically we used bands centered at 412 nm, 443 nm and 555 nm for the DOM absorption, chlorophyll maximum absorption and for absorption of suspended solid material respectively. For the fluorescence-based algorithm we analyzed in-situ remote sensing reflectance $(R_{rs})$ data using baseline method. Fluorescence Line Height $({\Delta}Flu)$ calculated from $R_{rs}$ at bands centered on 681 nm and 688 nm, and ${\Delta}Flu_{(area)}$ are used for development of algorithm. As a result ${\Delta}Flu_{(area)}$ method leads the best fitting for squared correlation coefficient $(R^2)$.

• Proceedings of the Korea Contents Association Conference
• /
• 2014.06a
• /
• pp.83-84
• /
• 2014

For hydrologic applications, lake levels is very important. As a first step in developing a remote-sensing based approach, lake stage estimation using remote sensing was proposed with the SRTM data from February 2000, which was providing a one-time snapshot. After several steps using contouring, masking, and CED, it was found that iterative contour fitting to a lake outline provided the outstanding result with the operator's decision. If the lake size is large enough, a constant meter of the difference removal due to bias found by Bhang et al. (2007) might be useful for more accurate estimations for the methods. A lake-level snapshot using SRTM data could provide estimates within 0.5 m level of accuracy for large lakes (> $10km^2$) with contouring. Also, even if the processing algorithm is complex, the accuracy was reliable. Overall, we confirmed that this study would provide useful information to ameliorate the quality of the SAR-derived DEMs specifically for water areas and if more expanded, SAR images can fruit result in water monitoring.

• Proceedings of the Korean Society of Machine Tool Engineers Conference
• /
• 2003.04a
• /
• pp.234-239
• /
• 2003

A cylindrical shape object is widely used as a mechanical part and a water pipe or an oil pipeline which are of cylindrical shape are widely used in the infrastructure. In order to handling such objects automatically using a robot, the posture i.e. orientation in 3D space should be recognized. However, since there is no edge or vertex in the pipe, it is very difficult task for the robot. In this paper in order to guide the robot, two kind of algorithms which find the axis using the measured range data from the robot to the object surface are to be developed. The algorithms are verified using both the simulated range data and the measured one.

• Communications for Statistical Applications and Methods
• /
• v.13 no.2
• /
• pp.233-241
• /
• 2006

Classical principal component analysis (PCA) can be formulated as finding the linear subspace that best accommodates multidimensional data points in the sense that the sum of squared residual distances is minimized. As alternatives to such LS (least squares) fitting approach, we produce LMS (least median of squares) and LTS (least trimmed squares)-type PCA by minimizing the median of squared residual distances and the trimmed sum of squares, in a similar fashion to Rousseeuw (1984)'s alternative approaches to LS linear regression. Proposed methods adopt the data-driven optimization algorithm of Croux and Ruiz-Gazen (1996, 2005) that is conceptually simple and computationally practical. Numerical examples are given.

• Journal of Korean Society for Geospatial Information Science
• /
• v.13 no.3 s.33
• /
• pp.87-94
• /
• 2005

3D positional information of lane can be automatically calculated tv combining GPS data, IMU data if coordinates of lane centers are given. The Road Safety Survey and Analysis Vehicle(RoSSAV) is currently under development to analyze three dimensional safety and stability of roads. RoSSAV has GPS and IMU sensors to get positional information of the vehicle and two road-facing CCD cameras for extraction of lane coordinates. In this paper, we develop technology that automatically detects centers of lanes from the road-facing cameras of RoSSAV. The proposed algorithm defines line-support regions by grouping pixels with similar edge orientation and magnitude together and extracts a line from each line support region by planar fitting. Then if extracted lines and the region in-between satisfy the criteria of brightness and width, we decide this region as lane. The proposed algorithm was more precise and stable than the previously proposed algorithm based on brightness threshold method. Experiments with real road scenes confirmed that lane was effectively extracted by the proposed algorithm.

• Journal of KIISE:Computer Systems and Theory
• /
• v.36 no.6
• /
• pp.511-520
• /
• 2009

In this paper, we present a new iso-density surface reconstruction scheme based on a hierarchy on the input volume data and the output mesh data. From the input volume data, we construct a hierarchy of volumes, called a volume pyramid, based on a 3D dilation filter. After constructing the volume pyramid, we extract a coarse base mesh from the coarsest resolution of the pyramid with the Cell-boundary representation scheme. We iteratively fit this mesh to the iso-points extracted from the volume data under O(3)-adjacency constraint. For the surface fitting, the shrinking process and the smoothing process are adopted as in the SWIS (Shrink-wrapped isosurface) algorithm[6], and we subdivide the mesh to be able to reconstruct fine detail of the isosurface. The advantage of our method is that it generates a mesh which can be utilized by several multiresolution algorithms such as compression and progressive transmission.