• 한국전자통신학회논문지
• /
• 제8권4호
• /
• pp.561-567
• /
• 2013

음성 인식을 위해 추출되는 특징벡터 중 일부는 드물게 나타나는 특이 패턴이다. 이들은 음성인식 시스템의 훈련에서 파라미터의 과도맞춤을 일으키며, 그 결과 새로운 입력 패턴의 인식을 저해하는 구조적 위험을 초래한다. 본 논문에서는 이러한 특이 패턴을 제거하는 하나의 방법으로서, 어느 크기 이상의 벡터를 제외시켜 음성인식 시스템의 훈련을 수행하는 방법에 대해 연구한다. 본 연구의 목적은 인식률을 저해시키지 않는 한도에서 가장 많은 특이 특징벡터를 제외시키는 것이다. 이를 위하여 우리는 하나의 절단 파라미터를 도입하고, 그 값의 변화가 FVQ(Fuzzy Vector Quantization)/HMM(Hidden Markov Model)을 사용한 화자독립 음성 인식에 미치는 영향을 조사하였다. 실험 결과, 인식률을 저하시키지 않는 특이 특징벡터의 수가 3%~6% 정도임을 확인하였다.

• 한국CDE학회논문집
• /
• 제15권3호
• /
• pp.243-252
• /
• 2010

This paper describes a new practical simplification method for feature-based solid models. In this approach, a solid model created using feature modeling operations is first simplified by the suppression of detailed features, and then, if necessary, the model is converted to a surface model to facilitate its modification. Finally, the simplified surface model is delivered to analysis packages. The algorithm was implemented based on CATIA V.5 and applied to mid-surface generation of plastic parts for structural analysis to prove the validity and usefulness.

• 한국정보과학회논문지:소프트웨어및응용
• /
• 제31권11호
• /
• pp.1534-1542
• /
• 2004

본 논문에서는 교통 영상에서 실시간 교통 정보를 산출하는 새로운 기법을 소개한다. 각 차선의 검지 영역은 통계적 특징과 형상적 특징을 이용하여 도로, 차량, 그리고 그림자 영역으로 분류한다. 한 프레임에서의 오류는 연속된 프레임에서의 차량 영역의 상관적 특징을 이용하여 시공간 영상에서 교정된다. 국부 검지 영역만을 처리하므로 전용의 병렬 처리기 없이도 초당 30 프레임 이상의 실시간 처리가 가능하며 기상조건, 그림자, 교통량의 변화에도 강건한 성능을 보장할 수 있다.

• 조명전기설비학회논문지
• /
• 제24권11호
• /
• pp.30-41
• /
• 2010

In this study we propose a new texture feature extraction method based on an estimation of the brightness and structural uniformity of CT images representing the important characteristics for emphysema recognition. The Center-Symmetric Local Binary Pattern (CS-LBP) is first used to combine gray level in order to describe the brightness uniformity characteristics of the CT image. Then the gradient orientation difference is proposed to generate another CS-LBP code combining with gray level to represent the structural uniformity characteristics of the CT image. The usage of the gray level, CS-LBP and gradient orientation differences enables the proposed method to extract rich and distinctive information from the CT images in multiple directions. Experimental results showed that the performance of the proposed method is more stable with respect to sensitivity and specificity when compared with the SGLDM, GLRLM and GLDM. The proposed method outperformed these three conventional methods (SGLDM, GLRLM, and GLDM) 7.85[%], 22.87[%], and 16.67[%] respectively, according to the diagnosis of average accuracy, demonstrated by the Receiver Operating Characteristic (ROC) curves.

• Smart Structures and Systems
• /
• 제29권1호
• /
• pp.153-166
• /
• 2022

Identifying fine cracks in steel bridge facilities is a challenging task of structural health monitoring (SHM). This study proposed an end-to-end crack image segmentation framework based on a one-step Convolutional Neural Network (CNN) for pixel-level object recognition with high accuracy. To particularly address the challenges arising from small object detection in complex background, efforts were made in loss function selection aiming at sample imbalance and module modification in order to improve the generalization ability on complicated images. Specifically, loss functions were compared among alternatives including the Binary Cross Entropy (BCE), Focal, Tversky and Dice loss, with the last three specialized for biased sample distribution. Structural modifications with dilated convolution, Spatial Pyramid Pooling (SPP) and Feature Pyramid Network (FPN) were also performed to form a new backbone termed CrackDet. Models of various loss functions and feature extraction modules were trained on crack images and tested on full-scale images collected on steel box girders. The CNN model incorporated the classic U-Net as its backbone, and Dice loss as its loss function achieved the highest mean Intersection-over-Union (mIoU) of 0.7571 on full-scale pictures. In contrast, the best performance on cropped crack images was achieved by integrating CrackDet with Dice loss at a mIoU of 0.7670.

• International Journal of Aerospace System Engineering
• /
• 제8권2호
• /
• pp.1-3
• /
• 2021

This study proposed a specific preliminary structural design procedure of the main wing for a small scale WIG vehicle to meet the target weight of the system requirement. The high stiffness and strength Carbon-Epoxy material was used for lightness, and the foam sandwich type structure at the upper skin and the spar webs was adopted for improvement of structural stability. After structural design, wing joint part was designed. Through investigation on structural design result, design modification was performed. After design modification, even thought the designed wing weight was a little bit heavier than the target wing weight, the structural safety and stability of the final design feature was confirmed.

• Structural Engineering and Mechanics
• /
• 제63권2호
• /
• pp.237-249
• /
• 2017

The Self-Organization Feature Map as an unsupervised network is very widely used these days in engineering science. The applied network in this paper is the Self Organization Feature Map with constant weights which includes Kohonen Network. In this research, Reinforced Concrete Shear Wall buildings with different stories and heights are analyzed and a database consisting of measured fundamental periods and characteristics of 78 RC SW buildings is created. The input parameters of these buildings include number of stories, height, length, width, whereas the output parameter is the fundamental period. In addition, using Genetic Algorithm, the structure of the Self-Organization Feature Map algorithm is optimized with respect to the numbers of layers, numbers of nodes in hidden layers, type of transfer function and learning. Evaluation of the SOFM model was performed by comparing the obtained values to the measured values and values calculated by expressions given in building codes. Results show that the Self-Organization Feature Map, which is optimized by using Genetic Algorithm, has a higher capacity, flexibility and accuracy in predicting the fundamental period.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제17권8호
• /
• pp.2101-2123
• /
• 2023

Recent studies have shown that the neural network-based binary code similarity detection technology performs well in vulnerability mining, plagiarism detection, and malicious code analysis. However, existing cross-architecture methods still suffer from insufficient feature characterization and low discrimination accuracy. To address these issues, this paper proposes a cross-architecture binary function similarity detection method based on composite feature model (SDCFM). Firstly, the binary function is converted into vector representation according to the proposed composite feature model, which is composed of instruction statistical features, control flow graph structural features, and application program interface calling behavioral features. Then, the composite features are embedded by the proposed hierarchical embedding network based on a graph neural network. In which, the block-level features and the function-level features are processed separately and finally fused into the embedding. In addition, to make the trained model more accurate and stable, our method utilizes the embeddings of predecessor nodes to modify the node embedding in the iterative updating process of the graph neural network. To assess the effectiveness of composite feature model, we contrast SDCFM with the state of art method on benchmark datasets. The experimental results show that SDCFM has good performance both on the area under the curve in the binary function similarity detection task and the vulnerable candidate function ranking in vulnerability search task.

• 한국전산구조공학회:학술대회논문집
• /
• 한국전산구조공학회 2001년도 가을 학술발표회 논문집
• /
• pp.93-100
• /
• 2001

The three dimensional(3D) reconstruction from two dimensional(2D) image data is using in many fields such as RPD(Rapid Product Development) and reverse engineering. In this paper, the main step of 3D reconstruction is comprised of two steps : image processing step and B-spline surface approximation step. In the image processing step, feature points of each cross-section are obtained by means of several image processing technologies. In the B-spline surface approximation step, using the data of feature points obtained in the image processing step, the control points of B-spline surface are obtained, which are used for IGES file of 3D CAD model.

• 스마트미디어저널
• /
• 제5권3호
• /
• pp.30-34
• /
• 2016

Nowadays, there are a lot of Korean documents, which often need to be identified in one of printed or handwritten text. Early methods for the identification use structural features, which can be simple and easy to apply to text of a specific font, but its performance depends on the font type and characteristics of the text. Recently, the bag-of-words model has been used for the identification, which can be invariant to changes in font size, distortions or modifications to the text. The method based on bag-of-words model includes three steps: word segmentation using connected component grouping, feature extraction, and finally classification using SVM(Support Vector Machine). In this paper, bag-of-words model based method is proposed using SURF(Speeded Up Robust Feature) for the identification of machine printed and handwritten text in Korean documents. The experiment shows that the proposed method outperforms methods based on structural features.