• 한국ITS학회 논문지
• /
• 제20권4호
• /
• pp.95-105
• /
• 2021

도로 노면 파손 탐지는 쾌적한 주행 환경과 안전사고의 예방을 위해 필요하다. 도로 관리 기관은 자동화 기술 기반의 검사 장비와 시스템을 활용하고 있다. 이러한 자동화 기술 중에서도 도로 노면의 파손을 탐지하는 기술은 중요한 역할을 수행한다. 최근 들어 딥러닝을 이용한 기술에 대한 연구가 활발하게 진행 중이다. 이러한 딥러닝 기술 개발을 위해서는 도로 영상과 라벨 영상이 필요하다. 하지만 라벨 영상을 확보하기 위해서는 많은 시간과 노동력이 요구된다. 본 논문에서는 이러한 문제를 해결하기 위하여 준지도 학습 기법 중 하나인 적대적 학습 방법을 제안했다. 이를 구현하기 위해서 5,327장의 도로 영상과 1,327장의 라벨 영상을 사용하여 경량화 심층 신경망 모델을 학습했다. 그리고 이를 400장의 도로 영상으로 실험한 결과 80.54%의 mean intersection over union과 77.85%의 F1 score를 갖는 모델을 개발하였다. 결과적으로 라벨 영상 없이 도로 영상만을 학습에 추가하여 인식 성능을 향상시킬 수 있는 기술을 개발하였고, 향후 도로 노면 관리를 위한 기술로 활용되길 기대한다.

• 디지털콘텐츠학회 논문지
• /
• 제19권6호
• /
• pp.1197-1205
• /
• 2018

본 논문에서는 함정전투체계의 EOTS나 IRST에서 획득한 영상을 초고해상도 영상으로 복원한다. 저해상도에서 초고해상도의 영상을 생성하는 생성 모델과 이를 판별하는 판별 모델로 구성된 생성적 적대 신경망을 이용하고, 다양한 학습 파라미터의 변화를 통한 최적의 값을 제안한다. 실험에 사용되는 학습 파라미터는 crop size와 sub-pixel layer depth, 학습 이미지 종류로 구성되며, 평가는 일반적인 영상 품질 평가 지표에 추가적으로 특징점 추출 알고리즘을 함께 사용하였다. 그 결과, Crop size가 클수록, Sub-pixel layer depth가 깊을수록, 고해상도의 학습이미지를 사용할수록 더 좋은 품질의 영상을 생성한다.

• 한국공간구조학회논문집
• /
• 제21권4호
• /
• pp.39-48
• /
• 2021

In this paper, a GAN-based data augmentation method is proposed for topology optimization. In machine learning techniques, a total amount of dataset determines the accuracy and robustness of the trained neural network architectures, especially, supervised learning networks. Because the insufficient data tends to lead to overfitting or underfitting of the architectures, a data augmentation method is need to increase the amount of data for reducing overfitting when training a machine learning model. In this study, the Ganerative Adversarial Network (GAN) is used to augment the topology optimization dataset. The produced dataset has been compared with the original dataset.

• KEPCO Journal on Electric Power and Energy
• /
• 제8권2호
• /
• pp.137-141
• /
• 2022

To inspect the power facility faults using artificial intelligence, it need that improve the accuracy of the diagnostic model are required. Data augmentation skill using generative adversarial network (GAN) is one of the best ways to improve deep learning performance. GAN model can create realistic-looking fake images using two competitive learning networks such as discriminator and generator. In this study, we intend to verify the effectiveness of virtual data generation technology by including the fake image of power facility generated through GAN in the deep learning training set. The GAN-based fake image was created for damage of LP insulator, and ResNet based normal and defect classification model was developed to verify the effect. Through this, we analyzed the model accuracy according to the ratio of normal and defective training data.

• 대한치과기공학회지
• /
• 제42권1호
• /
• pp.1-8
• /
• 2020

Purpose: Among the virtual teeth generated by Deep Convolutional Generative Adversarial Networks (DCGAN), the optimal data was analyzed for the number of learning. Methods: We extracted 50 mandibular first molar occlusal surfaces and trained 4,000 epoch with DCGAN. The learning screen was saved every 50 times and evaluated on a Likert 5-point scale according to five classification criteria. Results were analyzed by one-way ANOVA and tukey HSD post hoc analysis (α = 0.05). Results: It was the highest with 83.90±6.32 in the number of group3 (2,050-3,000) learning and statistically significant in the group1 (50-1,000) and the group2 (1,050-2,000). Conclusion: Since there is a difference in the optimal virtual tooth generation according to the number of learning, it is necessary to analyze the learning frequency section in various ways.

• Journal of information and communication convergence engineering
• /
• 제21권2호
• /
• pp.139-144
• /
• 2023

Engineers prefer deep neural networks (DNNs) for solving computer vision problems. However, DNNs pose two major problems. First, neural networks require large amounts of well-labeled data for training. Second, the covariate shift problem is common in computer vision problems. Domain adaptation has been proposed to mitigate this problem. Recent work on adversarial-learning-based unsupervised domain adaptation (UDA) has explained transferability and enabled the model to learn robust features. Despite this advantage, current methods do not guarantee the distinguishability of the latent space unless they consider class-aware information of the target domain. Furthermore, source and target examples alone cannot efficiently extract domain-invariant features from the encoded spaces. To alleviate the problems of existing UDA methods, we propose the mixup regularization in adversarial discriminative domain adaptation (ADDA) method. We validated the effectiveness and generality of the proposed method by performing experiments under three adaptation scenarios: MNIST to USPS, SVHN to MNIST, and MNIST to MNIST-M.

• 한국정보처리학회:학술대회논문집
• /
• 한국정보처리학회 2022년도 추계학술발표대회
• /
• pp.402-404
• /
• 2022

Recent research has shown that deep learning models are vulnerable to adversarial attacks not only in the digital but also in the physical domain. This becomes very critical for applications that have a very high safety concern, such as self-driving cars. In this study, we propose a physical adversarial attack technique for one of the common tasks in self-driving cars, namely segmentation of the urban scene. Our method can create a texture on a wall so that it can be misclassified as a road. The demonstration of the technique on a state-of-the-art cityscape pretrained model shows a fairly high success rate, which should raise awareness of more potential attacks in self-driving cars.

• 정보보호학회논문지
• /
• 제31권6호
• /
• pp.1215-1225
• /
• 2021

인공지능 기술은 우수한 성능을 기반으로 다양한 분야에 적용되고 있지만 입력 데이터에 인간이 감지할 수 없는 적대적 섭동을 추가하여 인공지능 모델의 오작동을 유도하는 적대적 예제에 취약하다. 현재까지 적대적 예제에 대응하기 위한 방법은 세 가지 범주로 분류할 수 있다. (1) 모델 재학습 방법; (2) 입력 변환 방법; (3) 적대적 예제 탐지 방법. 이러한 적대적 예제에 대응하기 위한 방법은 끊임없이 등장하고 있지만 각 적대적 공격 유형을 분류하는 연구는 미비한 실정이다. 따라서, 본 논문에서는 차원 축소와 군집화 알고리즘을 활용한 적대적 공격 유형 분류 방법을 제안한다. 구체적으로, 제안하는 방법은 적대적 예시로부터 적대적 섭동을 추출하고 선형 판별 분석(LDA)를 통해 적대적 섭동의 차원을 축소한 후에 k-means 알고리즘으로 적대적 공격 유형 분류를 수행한다. MNIST 데이터셋과 CIFAR-10 데이터셋을 대상으로 한 실험을 통해, 제안하는 기법은 5개의 적대적 공격(FGSM, BIM, PGD, DeepFool, C&W)을 효율적으로 분류할 수 있으며, 적대적 예제에 대한 정상 입력을 알 수 없는 제한적인 상황에서도 우수한 분류 성능을 나타내는 것을 확인하였다.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제13권11호
• /
• pp.5427-5445
• /
• 2019

While deep neural networks have achieved remarkable performance in representation learning, a huge amount of labeled training data are usually required by supervised deep models such as convolutional neural networks. In this paper, we propose a new representation learning method, namely generative adversarial networks (GAN) based bagging deep convolutional autoencoders (GAN-BDCAE), which can map data to diverse hierarchical representations in an unsupervised fashion. To boost the size of training data, to train deep model and to aggregate diverse learning machines are the three principal avenues towards increasing the capabilities of representation learning of neural networks. We focus on combining those three techniques. To this aim, we adopt GAN for realistic unlabeled sample generation and bagging deep convolutional autoencoders (BDCAE) for robust feature learning. The proposed method improves the discriminative ability of learned feature embedding for solving subsequent pattern recognition problems. We evaluate our approach on three standard benchmarks and demonstrate the superiority of the proposed method compared to traditional unsupervised learning methods.

• KEPCO Journal on Electric Power and Energy
• /
• 제7권2호
• /
• pp.323-328
• /
• 2021

Diagnostic models are required. Data augmentation is one of the best ways to improve deep learning performance. Traditional augmentation techniques that modify image brightness or spatial information are difficult to achieve great results. To overcome this, a generative adversarial network (GAN) technology that generates virtual data to increase deep learning performance has emerged. GAN can create realistic-looking fake images by competitive learning two networks, a generator that creates fakes and a discriminator that determines whether images are real or fake made by the generator. GAN is being used in computer vision, IT solutions, and medical imaging fields. It is essential to secure additional learning data to advance deep learning-based fault diagnosis solutions in the power industry where facilities are strictly maintained more than other industries. In this paper, we propose a method for generating power facility images using GAN and a strategy for improving performance when only used a small amount of data. Finally, we analyze the performance of the augmented image to see if it could be utilized for the deep learning-based diagnosis system or not.