• 한국융합학회논문지
• /
• 제12권11호
• /
• pp.45-51
• /
• 2021

딥러닝의 발전으로 인하여 의미론적 분할 방법은 다양한 분야에서 연구되고 있다. 의료 영상 분석과 같이 정확성을 요구하는 분야에서 분할 정확도가 떨어지는 문제가 있다. 본 논문은 의미론적 분할 시 특징 손실을 최소화하기 위해 딥러닝 기반 분할 방법인 PSPNet을 개선하였다. 기존 딥러닝 기반의 분할 방법은 특징 추출 및 압축 과정에서 해상도가 낮아져 객체에 대한 특징 손실이 발생한다. 이러한 손실로 윤곽선이나 객체 내부 정보에 손실이 발생하여 객체 분류 시 정확도가 낮아지는 문제가 있다. 이러한 문제를 해결하기 위해 의미론적 분할 모델인 PSPNet을 개선하였다. 기존 PSPNet에 제안하는 multi scale attention을 추가하여 객체의 특징 손실을 방지하였다. 기존 PPM 모듈에 attention 방법을 적용하여 특징 정제 과정을 수행하였다. 불필요한 특징 정보를 억제함으로써 윤곽선 및 질감 정보가 개선되었다. 제안하는 방법은 Cityscapes 데이터 셋으로 학습하였으며, 정량적 평가를 위해 분할 지표인 MIoU를 사용하였다. 실험을 통해 기존 PSPNet 대비 분할 정확도가 약 1.5% 향상되었다.

• 한국감성과학회:학술대회논문집
• /
• 한국감성과학회 2000년도 춘계 학술대회 및 국제 감성공학 심포지움 논문집 Proceeding of the 2000 Spring Conference of KOSES and International Sensibility Ergonomics Symposium
• /
• pp.138-143
• /
• 2000

In order to investigate the relationship between fabric sound parameters and subjective sensation, each sound from 60 fabrics was recorded and analyzed by Fast Fourier transform. Level pressure of total sound (LPT), three coefficients (ARC, ARF, ARE) of auto regressive models, loudness (Z), and sharpness (Z) by Zwickers model were estimated as sound parameters. For subjective evaluation, seven sensation (softness, loudness, sharpness, clearness, roughness, highness, and pleasantness) was rated by both semantic differential scale (SDS) and free modulus magnitude estimation (FMME). As the results, the ARC values were positively proportional to both LPT and loudness (Z) values. In both of SDS and FMME, softness, clearness, and pleasantness were negatively correlated with loudness, sharpness, roughness, and highness. In regression models, softness and clearness by FMME were negatively affected by LPT뭉 ARC, while loudness, sharpness, roughness, and highness were positively expected. Regression models for pleasantness showed low values for R2.

• 한국항해항만학회지
• /
• 제28권9호
• /
• pp.821-825
• /
• 2004

The purpose of this study is the assessment of changes in the leisure patterns of users of the Eulsook-do before and after the estuary dike construction. The following survey research method was conducted to implement the study, sampling by age was carried out selectively and randomly alike. A total of 319 persons were chosen for final analysis, excluding questionnaires answered found to be inappropriate. To measure the image of the lower Nakdong River around the Eulsook-do, with advice sought from experts, researchers, through several pilot tests, developed a 24-item semantic differential scale(SDS) that has two bi-polar adjectives at each end Study of the pre- and post- construction images of the Eulsook-do located on the coast of Busan suggests the dike construction has brought about negative changes in the Eulsook-do's overall image. There can be two alternatives for improving the negative image: restoration and development.

• 한국의류산업학회지
• /
• 제12권1호
• /
• pp.1-9
• /
• 2010

A purpose of this study was to find out how the casual and formal style clothes of stripe pattern giving variety by pattern direction, pattern width, and contrast coloration have an effect on image of wearers. 432 stimuli were made and 2,800 testee evaluated them using semantic differential scale. As a result, five image dimensions were drawn as a factor of attractiveness, activeness, gracefulness, visibility, and tenderness. In consequence of analysing the image difference by stripe pattern and clothing style, the stripe pattern and clothing style affect image presentation as a significant clue. And besides, as a result of positioning stimuli by image, pattern direction, coloration, and tone combination were important clues that decide image. Consequently, clothing style, stripe pattern, and contrast coloration were made clear as an efficient parameter in image presentation of clothing wearers.

• 정보과학회 논문지
• /
• 제44권4호
• /
• pp.383-391
• /
• 2017

빅데이터 시대가 도래 하면서 시맨틱 데이터의 양이 빠른 속도로 증가하고 있다. 이러한 대용량 시맨틱 데이터에서 의미 있는 암묵적 정보를 추론하기 위해서 지식 사용자의 경험적 지식을 기반으로 작성된 SWRL(Semantic Web Rule Language) 규칙들을 활용하는 많은 연구가 진행되고 있다. 그러나 기존의 단일 노드의 추론 시스템들은 대용량 데이터 처리에 한계가 있고, 다중 노드 기반의 분산 추론 시스템들은 네트워크 셔플링으로 인해 성능이 저하되는 문제점들이 존재한다. 따라서 본 논문에서는 기존 시스템들의 한계를 극복하고 보다 효율적인 분산 추론 방법을 제안한다. 또한 네트워크 셔플링을 최소화 할 수 있는 데이터 파티셔닝 전략을 소개하고, 점증적 추론에서 사용되는 추가된 새로운 데이터의 선별과 추론 규칙의 순서결정으로 추론 과정을 최적화 할 수 있는 방법에 대해 설명한다. 제안하는 방법의 성능을 측적하기 위해 약 2억 트리플로 구성된 WiseKB 온톨로지와 84개의 사용자 정의 규칙을 이용한 실험에서 32.7분이 소요되었다. 또한 LUBM 벤치 마크 데이터를 이용한 실험에서 맵-리듀스 방식에 비해 최대 2배 높은 성능을 보였다.

• 한국방송∙미디어공학회:학술대회논문집
• /
• 한국방송공학회 2009년도 IWAIT
• /
• pp.178-183
• /
• 2009

Online image dictionary has become more and more popular in concepts cognition. However, for existing online systems, only very few images are manually picked to demonstrate the concepts. Currently, there is very little research found on automatically choosing large scale online images with the help of semantic analysis. In this paper, we propose a novel framework to utilize community-generated online multimedia content to visually illustrate certain concepts. Our proposed framework adapts various techniques, including the correlation analysis, semantic and visual clustering to produce sets of high quality, precise, diverse and representative images to visually translate a given concept. To make the best use of our results, a user interface is deployed, which displays the representative images according the latent semantic coherence. The objective and subjective evaluations show the feasibility and effectiveness of our approach.

• International journal of advanced smart convergence
• /
• 제12권2호
• /
• pp.34-46
• /
• 2023

A growing number of large scale knowledge graphs raises several issues how knowledge graph data can be organized, discovered, and integrated efficiently. We present a novel semantic-based mashup platform for contents convergence which consists of acquisition, RDF storage, ontology learning, and mashup subsystems. This platform servers a basis for developing other more sophisticated applications required in the area of knowledge big data. Moreover, this paper proposes an entity matching method using graph convolutional network techniques as a preliminary work for automatic classification and discovery on knowledge big data. Using real DBP15K and SRPRS datasets, the performance of our method is compared with some existing entity matching methods. The experimental results show that the proposed method outperforms existing methods due to its ability to increase accuracy and reduce training time.

• 정보교육학회논문지
• /
• 제20권5호
• /
• pp.527-534
• /
• 2016

본 연구는 소프트웨어에 대해 초등학생들이 가진 이미지를 의미분별법을 이용하여 분석하였다. 검사의 항목은 총 35쌍의 소프트웨어 이미지 형용사를 선정하여 요인분석을 통해 7개의 주요인으로 구분하고 초등학생을 대상으로 조사하였다. 성별 간 SW이미지를 분석한 결과 남학생보다는 여학생이 소프트웨어를 어렵고 복잡하게 여기며 느리고 갖고 싶지 않다는 인식을 보였다. 또한 소프트웨어에 대한 자기 인식의 분석 결과 소프트웨어에 대해 잘 안다고 인식할 경우 소프트웨어에 대한 긍정적인 용어를 선호하고 있었다. 학년간 분석에서 고학년이 될수록 소프트웨어를 어렵고 복잡하며 소프트웨어의 객관적인 특징에 대한 답변을 하였다. 본 연구를 통해 소프트웨어 교육이 정규교과로 안착되는데 필요한 방향을 안내하였다.

• 한국조경학회지
• /
• 제24권2호
• /
• pp.13-24
• /
• 1996

The purpose of this study is to suggest optimum level of the Ratio of Greenery within the frame of vision in the interior landscape design through the analysis of visual character and preference of the interior landscape. The concept of the RG was defined as the ratio of projected area of plants higher than 1.2M against background wall from eye level. 5 photography of Interior landscape space -- 10,20,30,40 and 50% of the RG-- were constructed by computer graphic techinques. Likert scale and semantic differential scale were used to analyse visual character and preference of the interior landscape space. The analysis results are as follows : 1. Factors that compose of the image of the interior landscape have been found to be the "evaluation ", "complexity". The mean value of semantic differential scale showed a significant difference according to RG. When the RG was 20~30%, "Harmony" and "stability" was strongly recognized than the other factors and the interior landscape made the interior space natural and intimate. But at the RG 40!50%, users felt "stuffy" and "too complex" by many plants, so it was not efficient the Interior Landscape. 2. The visual preference was significantly different according to RG. The mean value of the visual preference was increased as the RG was higher, it was the highest at the RG 30%. But over the RG 30%, the preference level was declined. 3. Based on these results, this study suggests that the optimum level of RG in the Interior Landscape is 30%.at the optimum level of RG in the Interior Landscape is 30%.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제14권11호
• /
• pp.4426-4442
• /
• 2020

Since scene segmentation is becoming a hot topic in the field of autonomous driving and medical image analysis, researchers are actively trying new methods to improve segmentation accuracy. At present, the main issues in image semantic segmentation are intra-class inconsistency and inter-class indistinction. From our analysis, the lack of global information as well as macroscopic discrimination on the object are the two main reasons. In this paper, we propose a Densely connected residual Attention network (DA-Res2Net) which consists of a dense residual network and channel attention guidance module to deal with these problems and improve the accuracy of image segmentation. Specifically, in order to make the extracted features equipped with stronger multi-scale characteristics, a densely connected residual network is proposed as a feature extractor. Furthermore, to improve the representativeness of each channel feature, we design a Channel-Attention-Guide module to make the model focusing on the high-level semantic features and low-level location features simultaneously. Experimental results show that the method achieves significant performance on various datasets. Compared to other state-of-the-art methods, the proposed method reaches the mean IOU accuracy of 83.2% on PASCAL VOC 2012 and 79.7% on Cityscapes dataset, respectively.