• 한국정보처리학회:학술대회논문집
• /
• 한국정보처리학회 2021년도 추계학술발표대회
• /
• pp.878-879
• /
• 2021

Most state-of-the-art CNNs for action recognition are based on a two-stream architecture: RGB frames stream represents the appearance and the optical flow stream interprets the motion of action. However, the cost of optical flow computation is very high and then it increases action recognition latency. We introduce a design strategy for action recognition inspired by a two-stream network and teacher-student architecture. There are two sub-networks in our neural networks, the optical flow sub-network as a teacher and the RGB frames sub-network as a student. In the training stage, we distill the feature from the teacher as a baseline to train student sub-network. In the test stage, we only use the student so that the latency reduces without computing optical flow. Our experiments show that its advantages over two-stream architecture in both speed and performance.

• 한국음향학회지
• /
• 제39권3호
• /
• pp.143-149
• /
• 2020

본 논문에서는 딥러닝(deep learning)을 이용하여 환경음 분류 시 전처리 단계에서 사용하는 특징 추출 방법이 콘볼루션 신경망의 분류 성능에 미치는 영향에 대해서 다루었다. 이를 위해 환경음 분류 연구에서 많이 사용되는 UrbanSound8K 데이터셋에서 멜 스펙트로그램(mel spectrogram), 로그 멜 스펙트로그램(log mel spectrogram), Mel Frequency Cepstral Coefficient(MFCC), 그리고 delta MFCC를 추출하고 각각을 3가지 분포로 스케일링하였다. 이 데이터를 이용하여 4 종의 콘볼루션 신경망과 이미지넷에서 좋은 성능을 보였던 VGG16과 MobileNetV2 신경망을 학습시킨 다음 오디오 특징과 스케일링 방법에 따른 인식률을 구하였다. 그 결과 인식률은 스케일링하지 않은 로그 멜 스펙트럼을 사용했을 때 가장 우수한 것으로 나타났다. 도출된 결과를 모든 오디오 인식 문제로 일반화하기는 힘들지만, Urbansound8K의 환경음이 포함된 오디오를 분류할 때는 유용하게 적용될 수 있을 것이다.

• 한국인공지능학회지
• /
• 제11권4호
• /
• pp.1-8
• /
• 2023

In this paper, we propose a black ice detection platform framework using Convolutional Neural Networks (CNNs). To overcome black ice problem, we introduce a real-time based early warning platform using CNN-based architecture, and furthermore, in order to enhance the accuracy of black ice detection, we apply a multi-scale dilation convolution feature fusion (MsDC-FF) technique. Then, we establish a specialized experimental platform by using a comprehensive dataset of thermal road black ice images for a training and evaluation purpose. Experimental results of a real-time black ice detection platform show the better performance of our proposed network model compared to conventional image segmentation models. Our proposed platform have achieved real-time segmentation of road black ice areas by deploying a road black ice area segmentation network on the edge device Jetson Nano devices. This approach in parallel using multi-scale dilated convolutions with different dilation rates had faster segmentation speeds due to its smaller model parameters. The proposed MsCD-FF Net(2) model had the fastest segmentation speed at 5.53 frame per second (FPS). Thereby encouraging safe driving for motorists and providing decision support for road surface management in the road traffic monitoring department.

• International Journal of Computer Science & Network Security
• /
• 제22권2호
• /
• pp.214-222
• /
• 2022

Convolutional Neural networks (CNNs) are a category of deep learning networks that have proven very effective in computer vision tasks such as image classification. Notwithstanding, not much has been seen in its use for remote sensing image classification in developing countries. This is majorly due to the scarcity of training data. Recently, transfer learning technique has successfully been used to develop state-of-the art models for remote sensing (RS) image classification tasks using training and testing data from well-known RS data repositories. However, the ability of such model to classify RS test data from a different dataset has not been sufficiently investigated. In this paper, we propose a deep CNN model that can classify RS test data from a dataset different from the training dataset. To achieve our objective, we first, re-trained a ResNet-50 model using EuroSAT, a large-scale RS dataset to develop a base model then we integrated Augmentation and Ensemble learning to improve its generalization ability. We further experimented on the ability of this model to classify a novel dataset (Nig_Images). The final classification results shows that our model achieves a 96% and 80% accuracy on EuroSAT and Nig_Images test data respectively. Adequate knowledge and usage of this framework is expected to encourage research and the usage of deep CNNs for land cover mapping in cases of lack of training data as obtainable in developing countries.

• Journal of information and communication convergence engineering
• /
• 제16권3호
• /
• pp.148-152
• /
• 2018

Adversarial attacks on artificial intelligence (AI) systems use adversarial examples to achieve the attack objective. Adversarial examples consist of slightly changed test data, causing AI systems to make false decisions on these examples. When used as a tool for attacking AI systems, this can lead to disastrous results. In this paper, we propose an ensemble of degraded convolutional neural network (CNN) modules, which is more robust to adversarial attacks than conventional CNNs. Each module is trained on degraded images. During testing, images are degraded using various degradation methods, and a final decision is made utilizing a one-hot encoding vector that is obtained by summing up all the output vectors of the modules. Experimental results show that the proposed ensemble network is more resilient to adversarial attacks than conventional networks, while the accuracies for normal images are similar.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제15권4호
• /
• pp.1246-1262
• /
• 2021

Recent studies have demonstrated the strong ability of deep convolutional neural networks (CNNs) to significantly boost the performance in single image super-resolution (SISR). The key concern is how to efficiently recover and utilize diverse information frequencies across multiple network layers, which is crucial to satisfying super-resolution image reconstructions. Hence, previous work made great efforts to potently incorporate hierarchical frequencies through various sophisticated architectures. Nevertheless, economical SISR also requires a capable structure design to balance between restoration accuracy and computational complexity, which is still a challenge for existing techniques. In this paper, we tackle this problem by proposing a competent architecture called Enhanced U-Net Network (EUN), which can yield ready-to-use features in miscellaneous frequencies and combine them comprehensively. In particular, the proposed building block for EUN is enhanced from U-Net, which can extract abundant information via multiple skip concatenations. The network configuration allows the pipeline to propagate information from lower layers to higher ones. Meanwhile, the block itself is committed to growing quite deep in layers, which empowers different types of information to spring from a single block. Furthermore, due to its strong advantage in distilling effective information, promising results are guaranteed with comparatively fewer filters. Comprehensive experiments manifest our model can achieve favorable performance over that of state-of-the-art methods, especially in terms of computational efficiency.

• International Journal of Computer Science & Network Security
• /
• 제23권7호
• /
• pp.39-48
• /
• 2023

Breast cancer screening makes extensive utilization of mammography. Even so, there has been a lot of debate with regards to this application's starting age as well as screening interval. The deep learning technique of transfer learning is employed for transferring the knowledge learnt from the source tasks to the target tasks. For the resolution of real-world problems, deep neural networks have demonstrated superior performance in comparison with the standard machine learning algorithms. The architecture of the deep neural networks has to be defined by taking into account the problem domain knowledge. Normally, this technique will consume a lot of time as well as computational resources. This work evaluated the efficacy of the deep learning neural network like Visual Geometry Group Network (VGG Net) Residual Network (Res Net), as well as inception network for classifying the mammograms. This work proposed optimization of ResNet with Teaching Learning Based Optimization (TLBO) algorithm's in order to predict breast cancers by means of mammogram images. The proposed TLBO-ResNet, an optimized ResNet with faster convergence ability when compared with other evolutionary methods for mammogram classification.

• 문화기술의 융합
• /
• 제8권6호
• /
• pp.867-871
• /
• 2022

최근 빠른 유행의 변화 속에서 디자인의 변화는 패션기업의 매출에 큰 영향을 미치기 때문에 기업들은 신제품디자인 선택에 신중할 수밖에 없다. 최근 인공지능 분야의 발달에 따라 패션시장에서도 소비자들의 선호도를 높이기 위해 다양한 기계학습을 많이 활용하고 있다. 우리는 선호도와 같은 추상적인 개념을 수치화함으로써 신제품 개발에 신뢰성을 높이는 부분에 기여하고자 한다. 이를 위해 3가지 적대적 생성 신경망(Generative adversial netwrok, GAN)을 통하여 기존에 없는 새로운 이미지를 생성하고, 미리 훈련된 합성곱 신경망(Convolution neural networkm, CNN)을 이용하여 선호도라는 추상적인 개념을 수치화시켜 비교하였다. 심층 컨볼루션 적대적 생성 신경망(Deep convolutional generative adversial netwrok, DCGAN), 점진적 성장 적대적 생성 신경망(Progressive growing generative adversial netwrok, PGGAN), 이중 판별기 적대적 생성 신경망(Dual Discriminator generative adversial netwrok, D2GAN)의 3가지 방법을 통해 새로운 이미지를 생성하였고, 판매량이 높았던 제품으로 훈련된 합성곱 신경망으로 유사도를 비교, 측정하였다. 측정된 유사도의 정도를 선호도로 간주하였으며 실험 결과 D2GAN이 DCGAN, PGGAN에 비해 상대적으로 높은 유사도를 보여주었다.

• 국제학술발표논문집
• /
• The 8th International Conference on Construction Engineering and Project Management
• /
• pp.455-462
• /
• 2020

Noise which is unwanted sound is a serious pollutant that can affect human health, as well as the working and living environment if exposed to humans. However, current noise management on the construction project is generally conducted after the noise exceeds the regulation standard, which increases the conflicts with inhabitants near the construction site and threats to the safety and productivity of construction workers. To overcome the limitations of the current noise management methods, the activities of construction equipment which is the main source of construction noise need to be managed throughout the construction period in real-time. Therefore, this paper proposed a framework for automatically detecting noise sources in construction sites in real-time based on convolutional neural networks (CNNs) according to the following four steps: (i) Step 1: Definition of the noise sources; (ii) Step 2: Data preparation; (iii) Step 3: Noise source classification using the audio CNN; and (iv) Step 4: Noise source detection using the visual CNN. The short-time Fourier transform (STFT) and temporal image processing are used to contain temporal features of the audio and visual data. In addition, the AlexNet and You Only Look Once v3 (YOLOv3) algorithms have been adopted to classify and detect the noise sources in real-time. As a result, the proposed framework is expected to immediately find construction activities as current noise sources on the video of the construction site. The proposed framework could be helpful for environmental construction managers to efficiently identify and control the noise by automatically detecting the noise sources among many activities carried out by various types of construction equipment. Thereby, not only conflicts between inhabitants and construction companies caused by construction noise can be prevented, but also the noise-related health risks and productivity degradation for construction workers and inhabitants near the construction site can be minimized.

• 말소리와 음성과학
• /
• 제15권4호
• /
• pp.71-80
• /
• 2023

본 연구는 말더듬 화자들의 음성 데이터를 기반으로 하여, 인공지능 기술을 활용한 말더듬 자동 식별 방법을 개발하는 것을 주목적으로 진행되었다. 특히, 한국어를 모국어로 하는 말더듬 화자들을 대상으로 CNN(convolutional neural network) 알고리즘을 활용한 식별기 모델을 개발하고자 하였다. 이를 위해 말더듬 성인 9명과 정상화자 9명을 대상으로 음성 데이터를 수집하고, Google Cloud STT(Speech-To-Text)를 활용하여 어절 단위로 자동 분할한 후 유창, 막힘, 연장, 반복 등의 라벨을 부여하였다. 또한 MFCCs(mel frequency cepstral coefficients)를 추출하여 CNN 알고리즘을 기반한 말더듬 자동 식별기 모델을 수립하고자 하였다. 연장의 경우 수집결과가 5건으로 나타나 식별기 모델에서 제외하였다. 검증 결과, 정확도는 0.96으로 나타났고, 분류성능인 F1-score는 '유창'은 1.00, '막힘'은 0.67, '반복'은 0.74로 나타났다. CNN 알고리즘을 기반한 말더듬 자동분류 식별기의 효과를 확인하였으나, 막힘 및 반복유형에서는 성능이 미흡한 것으로 나타났다. 향후 말더듬의 유형별 충분한 데이터 수집을 통해 추가적인 성능 검증이 필요함을 확인하였다. 향후 말더듬 화자의 발화 빅데이터 확보를 통해 보다 신뢰성 있는 말더듬 자동 식별 기술의 개발과 함께 이를 통한 좀 더 고도화된 평가 및 중재 관련 서비스가 창출되기를 기대해 본다.