• KSCI Review
• /
• v.15 no.1
• /
• pp.179-183
• /
• 2007

다시점에서의 다중 객체 추적은 여러 분야에서 연구되고 있다. 다시점 영상 추적은 두 객체가 서로 근접하면 하나로 인식하는 문제점을 가지고 있다. 이러한 문제를 해결하기 위한 하나의 방법으로 능동형태모델(active shape mode: ASM)을 들 수 있다 ASM은 훈련집합을 이용하여 다른 객체에 가려진 목표 객체를 추적할 수 있다. 본 논문에서는 겹쳐진 객체를 추적하기 위해 ASM 기반의 다시점 추적 알고리듬(Multi-view tracking using ASM: MVTA)에 대해서 제안한다. 제안된 추적 방법은 (i) 영상 획득, (ii) 객체 추출, (iii) 객체 추적, 그리고 (iv) 현재 형태의 업데이트, 4가지 단계로 나눌 수 있다. 첫 번째 단계에서는 여러 대의 카메라를 사용해서 다시점 영상을 획득하며, 두 번째 단계에서는 객체를 배경으로부터 분리하며, 겹쳐진 객체로부터 목표 객체를 분리해낸다. 세 번째 단계에서는 추적을 위해 ASM을 사용하며, 마지막 단계인 네 번째 단계는 현재 입력 영상의 업데이트이다. 실험결과 제안한 MVTA는 겹쳐진 객체를 추적 시에 생기는 문제에 대해서 향상 된 결과를 보여준다.

• Annual Conference of KIPS
• /
• 2011.11a
• /
• pp.1229-1231
• /
• 2011

대량의 분류 규칙 탐사 과정은 앙상블기법을 사용하여 다양한 연구들이 이루어지고 있다. 본 논문에서는 의사결정 트리의 분열 문제와 singleton 포함 한계를 해결하기 위하여 Cascading-and-Sharing 앙상블 기법을 적용하여 점진적 다중 의사결정 트리를 구축하였다. 또한 분류의 정확도를 향상시키고, 트리의 복잡도와 모델 과잉접합을 피하기 위하여 다중 트리 구축과정에서 선형 상관분석기법을 기반으로 훈련 데이터 속성들의 중복성을 제거하였다. 실험 결과, 속성들의 중복성을 제거하여 구축한 트리들은 원래 기법보다 더 좋은 결과를 보여주었다.

• Annual Conference of KIPS
• /
• 2020.11a
• /
• pp.1052-1055
• /
• 2020

본 논문에서는 3D 메시 정보, RGB-D 손 자세 및 2D/3D 손/세그먼트 마스크를 포함하여 인간의 손과 관련된 다양한 컴퓨터 비전 작업에 사용할 수 있는 새로운 다중 모달 합성 벤치마크를 제안 하였다. 생성된 데이터셋은 기존의 대규모 데이터셋인 BigHand2.2M 데이터셋과 변형 가능한 3D 손 메시(mesh) MANO 모델을 활용하여 다양한 손 포즈 변형을 다룬다. 첫째, 중복되는 손자세를 줄이기 위해 전략적으로 샘플링하는 방법을 이용하고 3D 메시 모델을 샘플링된 손에 피팅한다. 3D 메시의 모양 및 시점 파라미터를 탐색하여 인간 손 이미지의 자연스러운 가변성을 처리한다. 마지막으로, 다중 모달리티 데이터를 생성한다. 손 관절, 모양 및 관점의 데이터 공간을 기존 벤치마크의 데이터 공간과 비교한다. 이 과정을 통해 제안된 벤치마크가 이전 작업의 차이를 메우고 있음을 보여주고, 또한 네트워크 훈련 과정에서 제안된 데이터를 사용하여 RGB 기반 손 포즈 추정 실험을 하여 생성된 데이터가 양질의 질과 양을 가짐을 보여준다. 제안된 데이터가 RGB 기반 3D 손 포즈 추정 및 시맨틱 손 세그멘테이션과 같은 품질 좋은 큰 데이터셋이 부족하여 방해되었던 작업에 대한 발전을 가속화할 것으로 기대된다.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2021.06a
• /
• pp.40-40
• /
• 2021

수재해 방지를 위한 수문해석 모형에서 정량적인 강수자료의 역할은 매우 중요하다. 최근에는 기후변화로 인한 국지성 집중호우 등 돌발 강수의 빈도가 증가하고 있어 지상에 설치된 우량계보다 시·공간적 변동성을 반영할 수 있는 격자형 위성 강수자료의 활용성이 커지고 있다. 하지만 위성강수자료는 관측 시에 대기의 상태 또는 위성별 관측 센서, 공간적 스케일 차이 등에 의해 실제 내린 강수와의 편의가 존재한다. 이를 해결하기 위해 지점 강수자료를 이용한 통계적, 지형정보학적 상세화 기법이 적용되고 있으나, 대부분의 연구에서 강수자료의 양적 보정만을 목적으로 수행되었다. 본 연구에서는 머신러닝 기반의 랜덤포레스트(random forest) 모델을 사용하여 다중위성 강수자료(CHIRPSv2, CMORPH, GSMaP, TRMMv7)와 기상청에서 제공하는 AWS, ASOS 지점 강수를 사용하여 최적 위성강수자료를 생성 후 각 위성강수자료와 비교·분석하였다. 2003년에서 2017년까지의 각 위성강수자료를 수집하여 같은 공간 스케일로 전처리한 뒤 모델에 입력하였으며 AWS 강수자료는 훈련, ASOS 강수자료는 검증에 이용되었다. 그 결과, 생성된 최적 위성강수자료는 각 위성강수자료보다 지점강수와의 편의가 줄고 높은 상관관계를 나타내고 있다. 이는 앞으로 사용될 위성강수자료의 시·공간적 보정 및 단기예측에 활용할 수 있으며, 특히 원격탐사자료의 의존도가 높은 미계측 대유역 수문해석에 정량적인 강수자료를 제공할 수 있을 것으로 판단된다.

• Smart Media Journal
• /
• v.12 no.9
• /
• pp.81-88
• /
• 2023

Recently, communication through online is increasing due to the spread of non-face-to-face services due to COVID-19. In non-face-to-face situations, the other person's opinions and emotions are recognized through modalities such as text, speech, and images. Currently, research on multimodal emotion recognition that combines various modalities is actively underway. Among them, emotion recognition using speech data is attracting attention as a means of understanding emotions through sound and language information, but most of the time, emotions are recognized using a single speech feature value. However, because a variety of emotions exist in a complex manner in a conversation, a method for recognizing multiple emotions is needed. Therefore, in this paper, we propose a multi-emotion regression model that extracts feature vectors after preprocessing speech data to recognize complex, inherent emotions and takes into account the passage of time.

• Convergence Security Journal
• /
• v.23 no.1
• /
• pp.117-123
• /
• 2023

To detect advanced ransomware attacks with machine learning-based models, the classification model must train learning data with high-dimensional feature space. And in this case, a 'curse of dimension' phenomenon is likely to occur. Therefore, dimensionality reduction of features must be preceded in order to increase the accuracy of the learning model and improve the execution speed while avoiding the 'curse of dimension' phenomenon. In this paper, we conducted classification of ransomware by applying three machine learning models and two feature extraction techniques to two datasets with extremely different dimensions of feature space. As a result of the experiment, the feature dimensionality reduction techniques did not significantly affect the performance improvement in binary classification, and it was the same even when the dimension of featurespace was small in multi-class clasification. However, when the dataset had high-dimensional feature space, LDA(Linear Discriminant Analysis) showed quite excellent performance.

• Korean Journal of Remote Sensing
• /
• v.33 no.2
• /
• pp.135-147
• /
• 2017

We, for the first time, estimated daily and monthly surface nitrogen dioxide ($NO_2$) volume mixing ratio (VMR) using three regression models with $NO_2$ tropospheric vertical column density (OMIT-rop $NO_2$ VCD) data obtained from Ozone Monitoring Instrument (OMI) in Seoul in South Korea at OMI overpass time (13:45 local time). First linear regression model (M1) is a linear regression equation between OMI-Trop $NO_2$ VCD and in situ $NO_2$ VMR, whereas second linear regression model (M2) incorporates boundary layer height (BLH), temperature, and pressure obtained from Atmospheric Infrared Sounder (AIRS) and OMI-Trop $NO_2$ VCD. Last models (M3M & M3D) are a multiple linear regression equations which include OMI-Trop $NO_2$ VCD, BLH and various meteorological data. In this study, we determined three types of regression models for the training period between 2009 and 2011, and the performance of those regression models was evaluated via comparison with the surface $NO_2$ VMR data obtained from in situ measurements (in situ $NO_2$ VMR) in 2012. The monthly mean surface $NO_2$ VMRs estimated by M3M showed good agreements with those of in situ measurements(avg. R = 0.77). In terms of the daily (13:45LT) $NO_2$ estimation, the highest correlations were found between the daily surface $NO_2$ VMRs estimated by M3D and in-situ $NO_2$ VMRs (avg. R = 0.55). The estimated surface $NO_2$ VMRs by three modelstend to be underestimated. We also discussed the performance of these empirical modelsfor surface $NO_2$ VMR estimation with respect to otherstatistical data such asroot mean square error (RMSE), mean bias, mean absolute error (MAE), and percent difference. This present study shows a possibility of estimating surface $NO_2$ VMR using the satellite measurement.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.31 no.6
• /
• pp.1171-1179
• /
• 2021

Image forensic is performed to check image limpidness. In this paper, a robust scheme is discussed to detect median filtering in low quality images. Detection of median filtering assists in overall image forensic. Improved spatial statistical features are extracted from the image to classify pristine and median filtered images. Image array data is rescaled to enhance the spatial statistical information. Features are extracted using Markov model on enhanced spatial statistics. Multiple difference arrays are considered in different directions for robust feature set. Further, texture operator features are combined to increase the detection accuracy and SVM binary classifier is applied to train the classification model. Experimental results are promising for images of low quality JPEG compression.

• The Journal of Bigdata
• /
• v.9 no.1
• /
• pp.153-168
• /
• 2024

In summer, as chillers are considered the main energy consumer of building, the efficient chiller operation is considered important. However, it is difficult to operate chillers to meet the cooling demand of the building as the demand fluctuates with various factors like the internal, external environment and behavior of the occupants and as chiller's constraint cause the current operation constrains operation in future. To address these problems, this study proposes a multi-chiller operation model based on deep reinforcement learning considering the minimum up-time of the chiller. The proposed model learns the value of the chiller operations according to the state composed of metrological and cooling system information and determines operation that minimizes the difference between the supply load and the cooling demand among feasible operations. The practical applicability was improved by applying the training algorithm considering the minimum up-time constraint and Experiments results using the actual data from a Korean university confirmed that the proposed model complies with the chiller constraints and outperforms the existing chiller operation logic of the university in terms of differences from the building cooling demand.

• 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.