• Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.35 no.5
• /
• pp.423-430
• /
• 2017

Recently, data analysis research has been carried out using the deep learning technique in various fields such as image interpretation and/or classification. Various types of algorithms are being developed for many applications. In this paper, we propose a precipitation prediction algorithm based on deep learning with high accuracy in order to take care of the possible severe damage caused by climate change. Since the geographical and seasonal characteristics of Korea are clearly distinct, the meteorological factors have repetitive patterns in a time series. Since the LSTM (Long Short-Term Memory) is a powerful algorithm for consecutive data, it was used to predict precipitation in this study. For the numerical test, we calculated the PWV (Precipitable Water Vapor) based on the tropospheric delay of the GNSS (Global Navigation Satellite System) signals, and then applied the deep learning technique to the precipitation prediction. The GNSS data was processed by scientific software with the troposphere model of Saastamoinen and the Niell mapping function. The RMSE (Root Mean Squared Error) of the precipitation prediction based on LSTM performs better than that of ANN (Artificial Neural Network). By adding GNSS-based PWV as a feature, the over-fitting that is a latent problem of deep learning was prevented considerably as discussed in this study.

• Journal of Communications and Networks
• /
• v.8 no.1
• /
• pp.93-105
• /
• 2006

The domination of the Internet by TCP-based services has spawned many efforts to provide high network utilization with low loss and delay in a simple and scalable manner. Active queue management (AQM) algorithms attempt to achieve these goals by regulating queues at bottleneck links to provide useful feedback to TCP sources. While many AQM algorithms have been proposed, most suffer from instability, require careful configuration of nonintuitive control parameters, or are not practical because of slow response to dynamic traffic changes. In this paper, we propose a new AQM algorithm, hybrid random early detection (HRED), that combines the more effective elements of recent algorithms with a random early detection (RED) core. HRED maps instantaneous queue length to a drop probability, automatically adjusting the slope and intercept of the mapping function to account for changes in traffic load and to keep queue length within the desired operating range. We demonstrate that straightforward selection of HRED parameters results in stable operation under steady load and rapid adaptation to changes in load. Simulation and implementation tests confirm this stability, and indicate that overall performances of HRED are substantially better than those of earlier AQM algorithms. Finally, HRED control parameters provide several intuitive approaches to trading between required memory, queue stability, and response time.

• Proceedings of the KSRS Conference
• /
• v.1
• /
• pp.417-420
• /
• 2006

In these days, mobile application dealing with information contents on mobile or handheld devices such as mobile communicator, PDA or WAP device face the most important industrial needs. The motivation of this study is the design and implementation of mobile application using high resolution satellite imagery, large-sized image data set. Although major advantages of mobile devices are portability and mobility to users, limited system resources such as small-sized memory, slow CPU, low power and small screen size are the main obstacles to developers who should handle a large volume of geo-based 3D model. Related to this, the previous works have been concentrated on GIS-based location awareness services on mobile; however, the mobile 3D terrain model, which aims at this study, with the source data of DEM (Digital Elevation Model) and high resolution satellite imagery is not considered yet, in the other mobile systems. The main functions of 3D graphic processing or pixel pipeline in this prototype are implemented with OpenGL|ES (Embedded System) standard API (Application Programming Interface) released by Khronos group. In the developing stage, experiments to investigate optimal operation environment and good performance are carried out: TIN-based vertex generation with regular elevation data, image tiling, and image-vertex texturing, text processing of Unicode type and ASCII type.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.15 no.1
• /
• pp.75-80
• /
• 2005

In this paper, a 3-dimensional analysis of magnetic road and a position sensing system for an autonomous vehicle system is described. Especially, a new position sensing system, end of the important component of an autonomous vehicle, is proposed. In a magnet based autonomous vehicle system, to sense the vehicle position, the sensor measures the field of magnetic road. The field depends on the sensor position of the vehicle on the magnetic road. As the rotation between the magnetic field and the sensor position is highly complex, it is difficult that the relation is stored in memory. Thus, a neural network is used to learn the mapping from th field to the position. The autonomous vehicle system with the proposed position sensing system is tested in experimental setup.

• Spatial Information Research
• /
• v.2 no.2
• /
• pp.107-116
• /
• 1994

Choropleth map is a type of thematic maps in which areal units are shaded with a color or pattern that symbolizes some characteristic of the mapped unit. CMS was first developed to produce choropleth maps on ordinary microcomputer environments in 1988. Since then there have been significant technological developments and enhancements in user environments, which have affected the field of choropleth mapping systems posi¬tively. A new version of CMS was developed in accordance with these changes. CMS II requires an IBM PC, or compatible, with the minimum 640KB memory and VGA graphic board. It supports HP laser jet printers to output a high resolution map. The program can use Hangul letters for main menu, map title, and legend. And dBase file format (DEW) was implemented to exchange attribute files effectively.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.26 no.8B
• /
• pp.1119-1128
• /
• 2001

본 논문에서는 이차원(2-D) 이산 웨이블릿 면환(Discrete Wavelet Transform, DWT)을 이용한 연상압축기를 FPGA 칩에서 실시간으로 동작 가능하도록 하는 효율적인 메모리 스케줄링 방법(E$^2$M$^2$)을 제안하였다. S/W적으로 위의 메모리 사상 방법을 검증한 후, 실제로 상용화된 SFRAM을 선정하여 메모리 제어기를 구현하였다. 본 논문에서는 Mallet-tree를 이용한 2-D DWT 영상압축 칩을 구현할 경우를 가정하였다. 이 알고리즘은 연산 과정에서 많은 데이터를 정장하여야 하는데, FPGA는 많은 데이터를 저장할 수 있는 메모리가 내장되어 있지 않으므로 외부 메모리를 사용하여야 한다. 외부메모리는 열(row)에 대해서만 연속(burst) 읽기, 쓰기 동작이 가능하기 때문에 Mallet-tree 알고리즘의 데이터 입출력을 그대로 적용할 경우 실시간 동작을 수행하는 DWT 압축 칩을 구현할 수 없다. 본 논문에서는 데이터 쓰기를 수행할 경우에는 메모리 셀(cell)의 수직 방향을 저장시키고 읽기를 수행할 때는 수평으로 데이터의 연속 읽기를 수행함으로써 필터가 항상 수평 방향에 위치하게 하는 방법을 제안하였다. 입방법을 C-언어로 DWT 커넬(Kernel)과 메모리의 에뮬레이터(emulator)를 구현하여 실험한 결과, Mallat-tree 이론을 그대로 적용시켰을 때와 동일한 필터링을 수행할 수 있음을 검증하였다. 또한, 상용화된 SDRAM의 메모리 제어기를 H/W로 구현하여 시뮬레이션 함으로써 본 논문에서 제안한 방법이 실제적인 하드웨어로 실시간 동작을 할 수 있음을 보였다.

• Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.28 no.2
• /
• pp.233-237
• /
• 2010

The recent advance of sensor networks and ubiquitous techniques allow collecting and analyzing of the data which overcome the limitation imposed by time and space in real-time for making decisions. Also, analysis and prediction of collected data can support useful and necessary information to users. The collected data in sensor networks environment is the stream data which has continuous, unlimited and sequential properties. Because of the continuous, unlimited and large volume properties of stream data, managing stream data is difficult. And the stream data needs dynamic processing method because of the memory constraint and access limitation. Accordingly, we analyze correlation stream data using principal component analysis. And using result of analysis, it helps users for making decisions.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.12 no.9
• /
• pp.4412-4428
• /
• 2018

The purpose of this work is to solve the problem of representing an entire video using Convolutional Neural Network (CNN) features for human action recognition. Recently, due to insufficient GPU memory, it has been difficult to take the whole video as the input of the CNN for end-to-end learning. A typical method is to use sampled video frames as inputs and corresponding labels as supervision. One major issue of this popular approach is that the local samples may not contain the information indicated by the global labels and sufficient motion information. To address this issue, we propose a binary hashing method to enhance the local feature extractors. First, we extract the local features and aggregate them into global features using maximum/minimum pooling. Second, we use the binary hashing method to capture the motion features. Finally, we concatenate the hashing features with global features using different normalization methods to train the classifier. Experimental results on the JHMDB and MPII-Cooking datasets show that, for these new local features, binary hashing mapping on the sparsely sampled features led to significant performance improvements.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.23 no.6
• /
• pp.1461-1470
• /
• 1998

Bit allocation or quantizer assigning problem is a basic and essential issue in lossy picture coding. It could be represented as minimizing overall distortion with the given constraint that total bits should not exceed allowed bit-budget. Optimal solution can be found by Lagrangian method. However this method needs much computational time and memory. This paper presents an approximation method that uses the activity measure. The comparison between the existing activity measuring techniques are made, and mapping function from activity value to the quantizer is proposed. Under MPEG-1 Intra coding situation, simulations show almost identical results compared to the optimal ones obtained by Lagrangian method with reduced computational time.

• Journal of Korea Society of Digital Industry and Information Management
• /
• v.9 no.1
• /
• pp.151-163
• /
• 2013

Recently the attention on digital hologram that is regarded as to be the final goal of the 3-dimensional video technology has been increased. Digital hologram is calculated by modeling the interference phenomenon between an object wave and a reference wave. The modeling for digital holograms is called by computer generated hologram (CGH) Generally, CGH requires a very large amount of calculation. So if holograms are generated in real time, high-speed method should be needed. In this paper, we analyzed CGH equation, optimized it for mapping general purpose graphic processing unit (GPGPU), and proposed a optimized CGH calculation technique for GPGPU by resource allocation and various experiments which include block size changing, memory selection, and hologram tiling. The implemented results showed that a digital hologram that has $1,024{\times}1,024$ resolution can be generated during approximately 24ms, using 1K point clouds. In the experiment, we used two GTX 580 GPGPU of nVidia Inc.