• Journal of Digital Contents Society
• /
• v.18 no.7
• /
• pp.1323-1331
• /
• 2017

Despite the fact that the copyright has grown into a large-scale business, there are many constant problems especially in image copyright. In this study, we propose an automatic object extraction and classification system for the scanned educational book image by combining document image processing and intelligent information technology like deep learning. First, the proposed technology removes noise component and then performs a visual attention assessment-based region separation. Then we carry out grouping operation based on extracted block areas and categorize each block as a picture or a character area. Finally, the caption area is extracted by searching around the classified picture area. As a result of the performance evaluation, it can be seen an average accuracy of 83% in the extraction of the image and caption area. For only image region detection, up-to 97% of accuracy is verified.

• Journal of Technology Innovation
• /
• v.29 no.4
• /
• pp.1-30
• /
• 2021

Due to the recent development of system semiconductors, technical innovation for the electric devices of the automobile industry is rapidly progressing. In particular, the electric device of automobiles is accelerating technology development competition among automobile parts makers, and the development cycle is also changing rapidly. Due to these changes, the importance of strategic planning for R&D is further strengthened. Due to the paradigm shift in the automobile industry, the Product-Technical Roadmap (P/TRM), one of the R&D strategies, analyzes technology forecasting, technology level evaluation, and technology acquisition method (Make/Collaborate/Buy) at the planning stage. The product-technical roadmap is a tool that identifies customer needs of products and technologies, selects technologies and sets development directions. However, most companies are developing the product-technical roadmap through a qualitative method that mainly relies on the technical papers, patent analysis, and expert Delphi method. In this study, empirical research was conducted through simulations that can supplement and strengthen the product-technical roadmap centered on the automobile industry by fusing Gartner's hype cycle, cumulative moving average-based data preprocessing, and deep learning (LSTM) time series analysis techniques. The empirical study presented in this paper can be used not only in the automobile industry but also in other manufacturing fields in general. In addition, from the corporate point of view, it is considered that it will become a foundation for moving forward as a leading company by providing products to the market in a timely manner through a more accurate product-technical roadmap, breaking away from the roadmap preparation method that has relied on qualitative methods.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.19 no.1
• /
• pp.317-326
• /
• 2024

Biometric recognition is a technology that determines whether a person is identified by extracting information on a person's biometric and behavioral characteristics with a specific device. Cyber threats such as forgery, duplication, and hacking of biometric characteristics are increasing in the field of biometrics. In response, the security system is strengthened and complex, and it is becoming difficult for individuals to use. To this end, multiple biometric models are being studied. Existing studies have suggested feature fusion methods, but comparisons between feature fusion methods are insufficient. Therefore, in this paper, we compared and evaluated the fusion method of multiple biometric models using fingerprint, face, and iris images. VGG-16, ResNet-50, EfficientNet-B1, EfficientNet-B4, EfficientNet-B7, and Inception-v3 were used for feature extraction, and the fusion methods of 'Sensor-Level', 'Feature-Level', 'Score-Level', and 'Rank-Level' were compared and evaluated for feature fusion. As a result of the comparative evaluation, the EfficientNet-B7 model showed 98.51% accuracy and high stability in the 'Feature-Level' fusion method. However, because the EfficietnNet-B7 model is large in size, model lightweight studies are needed for biocharacteristic fusion.

• Annual Conference of KIPS
• /
• 2020.05a
• /
• pp.446-450
• /
• 2020

딥러닝 기반의 영상 분석 방법들은 많은 양의 학습 데이터가 필요하며, 학습 데이터 구축에는 많은 시간과 노력이 소요된다. 특히 객체 검출 분야의 경우 영상 내 객체의 위치, 크기, 범주 등의 정보가 모두 필요하여 학습 데이터 구축에 더 많은 어려움이 있으며, 이를 해결하기 위해 최근 이미지 합성기반 데이터 증강에 관한 연구가 활발히 진행되고 있다. 이미지 합성기반 데이터 증강 방법은 배경 영상에 객체를 합성할 때 객체와 배경 영상이 접한 영역에서 아티팩트(Artifact)가 발생하며, 이는 객체 검출 모델이 아티팩트를 객체의 특징으로 모델링하여 검출 성능이 저하되는 원인이 된다. 이러한 문제를 해결하기 위하여 본 논문에서는 양방향 필터 기반의 이미지 합성 방법을 제안하고, 단일 단계 검출의 대표적인 방법인 RetinaNet을 이용하여 이미지 합성기반 데이터 증강 방법의 성능을 분석하였다. 공개 데이터셋에 대한 실험 결과 본 논문에서 사용한 단일 검출 방법 및 데이터 증강 기법을 사용하면 더 적은 양의 증강 데이터로 기존 방법과 동일한 성능을 보여주는 것을 확인하였다.

• Annual Conference on Human and Language Technology
• /
• 2019.10a
• /
• pp.362-367
• /
• 2019

본 논문에서는 Self-Attention을 활용한 딥러닝 기반 문맥의존 철자오류 교정 모델을 제안한다. 문맥의존 철자오류 교정은 최근 철자오류 교정 분야에서 활발히 연구되고 있는 문제 중 하나이다. 기존에는 규칙 기반, 확률 기반, 임베딩을 활용한 철자오류 교정이 연구되었으나, 아직 양질의 교정을 수행해내기에는 많은 문제점이 있다. 따라서 본 논문에서는 기존 교정 모델들의 단점을 보완하기 위해 Self-Attention을 활용한 문맥의존 철자오류 교정 모델을 제안한다. 제안 모델은 Self-Attention을 활용하여 기존의 임베딩 정보에 문맥 의존적 정보가 반영된 더 나은 임베딩을 생성하는 역할을 한다. 전체 문장의 정보가 반영된 새로운 임베딩을 활용하여 동적으로 타겟 단어와의 관련 단어들을 찾아 문맥의존 철자 오류교정을 시행한다. 본 논문에서는 성능평가를 위해 세종 말뭉치를 평가 데이터로 이용하여 제안 모델을 실험하였고, 비정형화된 구어체(Kakao Talk) 말뭉치로도 평가 데이터를 구축해 실험한 결과 비교 모델보다 높은 정확율과 재현율의 성능향상을 보였다.

• The Journal of Bigdata
• /
• v.6 no.2
• /
• pp.71-84
• /
• 2021

Since small and medium-sized enterprises fell short of the securement of technological competitiveness in the field of big data and artificial intelligence (AI) field-core technologies of the Fourth Industrial Revolution, it is important to strengthen the competitiveness of the overall industry through technology commercialization. In this study, we aimed to propose a priority related to technology transfer and commercialization for practical use of public research results. We utilized public research performance information, improving missing values of 6T classification by deep learning model with an ensemble method. Then, we conducted topic modeling to derive the converging fields of big data and AI. We classified the technology fields into four different segments in the technology portfolio based on technology activity and technology efficiency, estimating the potential of technology commercialization for those fields. We proposed a priority of technology commercialization for 10 detailed technology fields that require long-term investment. Through systematic analysis, active utilization of technology, and efficient technology transfer and commercialization can be promoted.

• Convergence Security Journal
• /
• v.21 no.3
• /
• pp.57-66
• /
• 2021

As the detection performance using deep learning and machine learning of the intrusion detection field has been verified, the cases of using it are increasing day by day. However, it is difficult to collect the data required for learning, and it is difficult to apply the machine learning performance to reality due to the imbalance of the collected data. Therefore, in this paper, A mixed sampling technique using t-SNE visualization for imbalanced data processing is proposed as a solution to this problem. To do this, separate fields according to characteristics for intrusion detection events, including payload. Extracts TF-IDF-based features for separated fields. After applying the mixed sampling technique based on the extracted features, a data set optimized for intrusion detection with imbalanced data is obtained through data visualization using t-SNE. Nine sampling techniques were applied through the open intrusion detection dataset CSIC2012, and it was verified that the proposed sampling technique improves detection performance through F-score and G-mean evaluation indicators.

• The Journal of the Convergence on Culture Technology
• /
• v.9 no.4
• /
• pp.687-692
• /
• 2023

We analyze the accuracy of emotion analysis of natural language processing deep learning model and propose to use it for emotional content development. After looking at the outline of the GPT-3 model, about 6,000 pieces of dialogue data provided by Aihub were input to 9 emotion categories: 'joy', 'sadness', 'fear', 'anger', 'disgust', and 'surprise'. ', 'interest', 'boredom', and 'pain'. Performance evaluation was conducted using the evaluation indices of accuracy, precision, recall, and F1-score, which are evaluation methods for natural language processing models. As a result of the emotion analysis, the accuracy was over 91%, and in the case of precision, 'fear' and 'pain' showed low values. In the case of reproducibility, a low value was shown in negative emotions, and in the case of 'disgust' in particular, an error appeared due to the lack of data. In the case of previous studies, emotion analysis was mainly used only for polarity analysis divided into positive, negative, and neutral, and there was a limitation in that it was used only in the feedback stage due to its nature. We expand emotion analysis into 9 categories and suggest its use in the development of emotional content considering it from the planning stage. It is expected that more accurate results can be obtained if emotion analysis is performed by additionally collecting more diverse daily conversations through follow-up research.

• The Journal of Bigdata
• /
• v.7 no.1
• /
• pp.89-97
• /
• 2022

Various efforts are needed to prevent accidents because ship collisions can cause various negative situations such as economic losses and casualties. Therefore, research to prevent accidents is being actively conducted, and in this study, new leading indicators for preventing ship collision accidents is proposed. In previous studies, the risk of collision was expressed in consideration of the distance between ships in a specific sea area, but there is a disadvantage that a new model needs to be developed to apply this to other sea areas. In this study, the density-based ship domain DESD (Density-based Empirical Ship Domain) including the environment and operating characteristics of the sea area was defined using AIS (Automatic Identification System) data, which is ship operation information. Deep clustering is applied to two-dimensional DESDs created for each sea area to cluster the seas with similar operating environments. Through the analysis of the relationship between clustered sea areas and ship collision accidents, it was statistically tested that the occurrence of accidents varies by characteristic of each sea area, and it was proved that DESD can be used as a leading indicator of accidents.

• Review of KIISC
• /
• v.29 no.6
• /
• pp.67-74
• /
• 2019

기계학습 알고리즘을 이용한 다양한 분야에서의 활용사례들이 우리 사회로 점차 확산되어가며, 기계학습을 통해 산출된 모델의 오동작을 유발할 수 있는 공격이 활발히 연구되고 있다. 특히, 한국에서는 딥러닝을 포함해 인공지능을 응용한 융합분야를 국가적 차원에서 추진하고 있으며, 만약 인공지능 모델 자체에서 발생하는 취약점을 보완하지 못하고 사전에 공격을 대비하지 않는다면, 뒤늦은 대응으로 인하여 관련 산업의 활성화가 지연될 수 있는 문제점이 발생할 수도 있다. 본 논문에서는 기계학습 모델에서, 특히 심층 신경망으로 구성되어 있는 모델에서 발생할 수 있는 공격들을 정의하고 연구 동향을 분석, 안전한 기계학습 모델 구성을 위해 필요한 시사점을 제시한다. 구체적으로, 가장 널리 알려진 적대적 사례(adversarial examples) 뿐 아니라, 프라이버시 침해를 유발하는 추론 공격 등이 어떻게 정의되는지 설명한다.