• Korean Journal of Remote Sensing
• /
• v.34 no.5
• /
• pp.811-827
• /
• 2018

The purpose of this study is to compare machine learning algorithm and deep learning algorithm in crop classification using multi-temporal remote sensing data. For this, impacts of machine learning and deep learning algorithms on (a) hyper-parameter and (2) training sample size were compared and analyzed for Haenam-gun, Korea and Illinois State, USA. In the comparison experiment, support vector machine (SVM) was applied as machine learning algorithm and convolutional neural network (CNN) was applied as deep learning algorithm. In particular, 2D-CNN considering 2-dimensional spatial information and 3D-CNN with extended time dimension from 2D-CNN were applied as CNN. As a result of the experiment, it was found that the hyper-parameter values of CNN, considering various hyper-parameter, defined in the two study areas were similar compared with SVM. Based on this result, although it takes much time to optimize the model in CNN, it is considered that it is possible to apply transfer learning that can extend optimized CNN model to other regions. Then, in the experiment results with various training sample size, the impact of that on CNN was larger than SVM. In particular, this impact was exaggerated in Illinois State with heterogeneous spatial patterns. In addition, the lowest classification performance of 3D-CNN was presented in Illinois State, which is considered to be due to over-fitting as complexity of the model. That is, the classification performance was relatively degraded due to heterogeneous patterns and noise effect of input data, although the training accuracy of 3D-CNN model was high. This result simply that a proper classification algorithms should be selected considering spatial characteristics of study areas. Also, a large amount of training samples is necessary to guarantee higher classification performance in CNN, particularly in 3D-CNN.

• Journal of KIISE:Computer Systems and Theory
• /
• v.29 no.6
• /
• pp.307-317
• /
• 2002

In this paper we propose a new mesh reconstruction scheme that produces a displaced subdivision surface directly from unorganized points. The displaced subdivision surface is a new mesh representation that defines a detailed mesh with a displacement map over a smooth domain surface, but original displaced subdivision surface algorithm needs an explicit polygonal mesh since it is not a mesh reconstruction algorithm but a mesh conversion (remeshing) algorithm. The main idea of our approach is that we sample surface detail from unorganized points without any topological information. For this, we predict a virtual triangular face from unorganized points for each sampling ray from a parameteric domain surface. Direct displaced subdivision surface reconstruction from unorganized points has much importance since the output of this algorithm has several important properties: It has compact mesh representation since most vertices can be represented by only a scalar value. Underlying structure of it is piecewise regular so it ran be easily transformed into a multiresolution mesh. Smoothness after mesh deformation is automatically preserved. We avoid time-consuming global energy optimization by employing the input data dependant mesh smoothing, so we can get a good quality displaced subdivision surface quickly.

• Journal of KIISE:Computer Systems and Theory
• /
• v.33 no.6
• /
• pp.354-362
• /
• 2006

This paper suggests a method for generating high qualify stereo video avatar to support visual communication in a CAVE$^{TM}$-like system. In such a system because of frequent change of light projected onto screens around user, it is not easy to extract user silhouette robustly, which is an essential step to generate a video avatar. In this study, we use an infrared reflective image acquired by a grayscale camera with a longpass filter so that the change of visible light on a screen is blocked to extract robust user silhouette. In addition, using two color cameras positioned at a distance of a binocular disparity of human eyes, we acquire two stereo images of the user for fast generation and stereoscopic display of a high quality video avatar without 3D reconstruction. We also suggest a fitting algorithm of a silhouette mask on an infrared reflective image into an acquired color image to remove background. Generated stereo images of a video avatar are texture mapped into a plane in virtual world and can be displayed in stereoscopic using frame sequential stereo method. Suggested method have advantages that it generates high quality video avatar taster than 3D approach and it gives stereoscopic feeling to a user 2D based approach can not provide.

• 제어로봇시스템학회:학술대회논문집
• /
• 1991.10b
• /
• pp.1639-1645
• /
• 1991

Because of the important role LD converters play in the production of high quality steel, various dynamic models have been attempted in the past by many researchers not only to understand the complex chemical reactions that take place in the converter process but also to assist the converter operation itself using computers. And yet no single dynamic model was found to be completely satisfactory because of the complexity involved with the process. The process indeed involves dynamic energy and mass balances at high temperatures accompanied by complex chemical reactions and transport phenomena in the molten state. In the present study, a mathematical model describing the dynamic behavior of LD converter process has been developed. The dynamic model describes the time behavior of the temperature and the concentrations of chemical species in the hot metal bath and slag. The analysis was greatly facilitated by dividing the entire process into three zones according to the physical boundaries and reaction mechanisms. These three zones were hot metal (zone 1), slag (zone 2) and emulsion (zone 3) zones. The removal rate of Si, C, Mn and P and the rate of Fe oxidation in the hot metal bath, and the change of composition in the slag were obtained as functions of time, operating conditions and kinetic parameters. The temperature behavior in the metal bath and the slag was also obtained by considering the heat transfer between the mixing and the slag zones and the heat generated from chemical reactions involving oxygen blowing. To identify the unknown parameters in the equations and simulate the dynamic model, Hooke and Jeeves parttern search and Runge-Kutta integration algorithm were used. By testing and fitting the model with the data obtained from the operation of POSCO #2 steelmaking plant, the dynamic model was able to predict the characteristics of the main components in the LD converter. It was possible to predict the optimum CO gas recovery by computer simulation

• KSCE Journal of Civil and Environmental Engineering Research
• /
• v.36 no.1
• /
• pp.105-114
• /
• 2016

With approximately 20 million transportation card data entries of the metropolitan districts being generated per day, application of the data to management and policy interventions is becoming an issue of interest. The research herein attempts a model of the possibility of dynamic demand change predictions and its purpose is thereby to construct a Dynamic Passengers Trip Assignment Model. The model and algorithm created are targeted at city rail lines operated by seven different transport facilities with the exclusion of travel by bus, as passenger movements by this mode can be minutely disaggregated through card tagging. The model created has been constructed in continuous time as is fitting to the big data characteristic of transport card data, while passenger path choice behavior is effectively represented using a perception parameter as a function of increasing number of transfers. Running the model on 800 pairs of metropolitan city rail data has proven its capability in determining dynamic demand at any moment in time, in line with the typical advantages expected of a continuous time-based model. Comparison against data measured by the eye of existing rail operating facilities to assess changes in congestion intensity shows that the model closely approximates the values and trends of the existing data with high levels of confidence. Future research efforts should be directed toward continued examination into construction of an integrated bus-city rail system model.

• Journal of Korean Society of Environmental Engineers
• /
• v.31 no.11
• /
• pp.975-982
• /
• 2009

The purpose of this study is to offer informations about the current conditions and basic data of stream sediments in Busan city. So we first select 14 urban streams and collect sediment samples. Then, COD, proximate analysis, volatile solid, organic carbon content and elemental analysis were conducted to determine characteristics of the sediments. Results show that COD, volatile solid, Organic carbon content, T-N of sediment are determined in the range of 1.20~75.07 mg/g, 0.19~11.54%, 0.23~34.21% and 0.76~3.46%, respectively. Analysis data of sediments were compared with USEPA sediment quality standards and ontario sediment quality guidelines. As a result, when compared with COD, volatile solid and organic carbon content values, Bosucheon and Gudeokcheon are relatively heavily contaminated than the remainder sampling sites. But when compared with T-N values, all of sites were evaluated as seriously contaminated. Finally, for the determination of the correlations between sediment COD and moisture contents, ash contents, volatile solid, total organic carbon, total nitrogen and total carbon, linear model was fitted to the data using a least-squares algorithm. As a result, Linear model was well fitted to each data with good values of the correlation coefficient (r=0.9664~0.8501).

• Asian-Australasian Journal of Animal Sciences
• /
• v.30 no.11
• /
• pp.1550-1556
• /
• 2017

Objective: This study estimated the genetic parameters for productive and reproductive traits. Methods: The data included production and reproduction records of animals that have calved between 1979 and 2013. The genetic parameters were estimated using multivariate mixed models (DMU) package, fitting univariate and multivariate mixed models with average information restricted maximum likelihood algorithm. Results: The estimates of heritability for milk production traits from the first three lactation records were $0.03{\pm}0.03$ for lactation length (LL), $0.17{\pm}0.04$ for lactation milk yield (LMY), and $0.15{\pm}0.04$ for 305 days milk yield (305-d MY). For reproductive traits the heritability estimates were, $0.09{\pm}0.03$ for days open (DO), $0.11{\pm}0.04$ for calving interval (CI), and $0.47{\pm}0.06$ for age at first calving (AFC). The repeatability estimates for production traits were $0.12{\pm}0.02$, for LL, $0.39{\pm}0.02$ for LMY, and $0.25{\pm}0.02$ for 305-d MY. For reproductive traits the estimates of repeatability were $0.19{\pm}0.02$ for DO, and to $0.23{\pm}0.02$ for CI. The phenotypic correlations between production and reproduction traits ranged from $0.08{\pm}0.04$ for LL and AFC to $0.42{\pm}0.02$ for LL and DO. The genetic correlation among production traits were generally high (>0.7) and between reproductive traits the estimates ranged from $0.06{\pm}0.13$ for AFC and DO to $0.99{\pm}0.01$ between CI and DO. Genetic correlations of productive traits with reproductive traits were ranged from -0.02 to 0.99. Conclusion: The high heritability estimates observed for AFC indicated that reasonable genetic improvement for this trait might be possible through selection. The $h^2$ and r estimates for reproductive traits were slightly different from single versus multi-trait analyses of reproductive traits with production traits. As single-trait method is biased due to selection on milk yield, a multi-trait evaluation of fertility with milk yield is recommended.