• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2019.06a
• /
• pp.5-8
• /
• 2019

단일 영상 초해상도 (Single Image Super-Resolution - SISR)기법은 카메라로 획득된 저해상도 영상에 필터 기반의 연산을 적용하여 좋은 화질의 고해상도 영상을 복원하는 과정이다. 최근에 심층 합성곱 신경망 학습의 발전에 따라 단일 영상 초해상도에 적용되는 심층 학습 기법들은 좋은 성과를 보여 주고 있다. 본 논문은 단일 영상 초해상도 성능을 개선하기 위해 웨이블릿 예측 네트워크를 효율적으로 적용하는 방법에 대해 연구하였으며, 저해상도 입력 영상의 특징을 잘 추출해내기 위해 네트워크 내부에 RDB를 적용하여 기존 방식보다 효율적으로 고해상도 영상 복원하는 기법을 제안한다. 모의실험을 통해 제안하는 방법이 기존 방법보다 화질은 약 PSNR 0.18dB만큼 우수하며 속도는 1.17배 빠른 것을 확인하였다.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.17 no.4
• /
• pp.1234-1257
• /
• 2023

In this paper, we propose a framework for multi-focus image fusion called PATN. In our approach, by aggregating deep features extracted based on the U-type Transformer mechanism and shallow features extracted using the PSA module, we make PATN feed both long-range image texture information and focus on local detail information of the image. Meanwhile, the edge-preserving information value of the fused image is enhanced using a dense residual block containing the Sobel gradient operator, and three loss functions are introduced to retain more source image texture information. PATN is compared with 17 more advanced MFIF methods on three datasets to verify the effectiveness and robustness of PATN.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2022.06a
• /
• pp.1322-1324
• /
• 2022

최근 딥러닝의 발전에 따라 단일 이미지 초해상도 분야에 좋은 성과를 보여주고 있다. 그러나 보다 더 높은 성능을 획득하기 위해 네트워크의 깊이 및 파라미터의 수가 크게 증가하였고, 모바일 및 엣지 디바이스에 원활하게 적용되기 위하여 딥러닝 모델 경량화의 필요성이 대두되고 있다. 이에 본 논문에서는 초해상도 모델 중 하나인 EDSR(Enhanced Deep Residual Network)에 대조 학습 기반 지식 전이를 적용한 경량화 기법을 제안한다. 실험 결과 제안한 지식 전이 기법이 기존의 다른 지식 증류 기법보다 향상된 성능을 보임을 확인하였다.

• Proceedings of the Korean Society of Computer Information Conference
• /
• 2023.07a
• /
• pp.519-520
• /
• 2023

본 논문에서는 최근 반려동물 등록제가 확대되고 있는 바, 기존의 마이크로 칩 삽입 방법을 회피하고 반려견 이미지를 통하여 개체를 인식하는 방법을 연구하였다. 반려견의 전체 이미지를 학습시켜 해당 개체를 식별하는 지능형 시스템을 ResNet 알고리즘을 이용하여 구현하고, 수집된 반려견의 개체 사진을 학습시켜 필요한 개체를 식별할 수 있도록 하였다.

• Korean Journal of Remote Sensing
• /
• v.38 no.1
• /
• pp.111-126
• /
• 2022

Land use and land cover (LULC) mapping is an important factor in geospatial analysis. Although highly precise ground-based LULC monitoring is possible, it is time consuming and costly. Conversely, because the synthetic aperture radar (SAR) sensor is an all-weather sensor with high resolution, it could replace field-based LULC monitoring systems with low cost and less time requirement. Thus, LULC is one of the major areas in SAR applications. We developed a LULC model using only KOMPSAT-5 single co-polarized data and digital elevation model (DEM) data. Twelve HH-polarized images and 18 VV-polarized images were collected, and two HH-polarized images and four VV-polarized images were selected for the model testing. To train the LULC model, we applied the conditional generative adversarial network (cGAN) method. We used U-Net combined with the residual unit (ResUNet) model to generate the cGAN method. When analyzing the training history at 1732 epochs, the ResUNet model showed a maximum overall accuracy (OA) of 93.89 and a Kappa coefficient of 0.91. The model exhibited high performance in the test datasets with an OA greater than 90. The model accurately distinguished water body areas and showed lower accuracy in wetlands than in the other LULC types. The effect of the DEM on the accuracy of LULC was analyzed. When assessing the accuracy with respect to the incidence angle, owing to the radar shadow caused by the side-looking system of the SAR sensor, the OA tended to decrease as the incidence angle increased. This study is the first to use only KOMPSAT-5 single co-polarized data and deep learning methods to demonstrate the possibility of high-performance LULC monitoring. This study contributes to Earth surface monitoring and the development of deep learning approaches using the KOMPSAT-5 data.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.25 no.12
• /
• pp.1853-1858
• /
• 2021

O-rings fill the gaps between mechanical parts. Until now, the sorting of defective products has been performed visually and manually, so classification errors often occur. Therefore, a camera-based defect classification system without human intervention is required. However, a binarization process is required to separate the required region from the background in the camera input image. In this paper, an adaptive binarization technique that considers the surrounding pixel values is applied to solve the problem that single-threshold binarization is difficult to apply due to factors such as changes in ambient lighting or reflections. In addition, the convex hull technique is also applied to compensate for the missing pixel part. And the learning model to be applied to the separated region applies the residual error-based deep learning neural network model, which is advantageous when the defective characteristic is non-linear. It is suggested that the proposed system through experiments can be applied to the automation of O-ring defect detection.

• Journal of IKEEE
• /
• v.28 no.2
• /
• pp.168-173
• /
• 2024

Speaker recognition refers to a technology that analyzes voice frequencies that are different for each individual and compares them with pre-stored voices to determine the identity of the person. Deep learning-based speaker recognition is being applied to many fields, and pet robots are one of them. However, the hardware performance of pet robots is very limited in terms of the large memory space and calculations of deep learning technology. This is an important problem that pet robots must solve in real-time interaction with users. Lightening deep learning models has become an important way to solve the above problems, and a lot of research is being done recently. In this paper, we describe the results of research on lightweight speaker recognition for pet robots by constructing a voice data set for pet robots, which is a specific command type, and comparing the results of models using residuals. In the conclusion, we present the results of the proposed method and Future research plans are described.

• Korean Journal of Remote Sensing
• /
• v.36 no.3
• /
• pp.449-460
• /
• 2020

Super-resolution is a technique used to reconstruct an image with low-resolution into that of high-resolution. Recently, deep-learning based super resolution has become the mainstream, and applications of these methods are widely used in the remote sensing field. In this paper, we propose a super-resolution method based on the deep back-projection network model to improve the satellite image resolution by the factor of four. In the process, we customized the loss function with the edge loss to result in a more detailed feature of the boundary of each object and to improve the stability of the model training using generative adversarial network based on Wasserstein distance loss. Also, we have applied the detail preserving image down-scaling method to enhance the naturalness of the training output. Finally, by including the modified-residual learning with a panchromatic feature in the final step of the training process. Our proposed method is able to reconstruct fine features and high frequency information. Comparing the results of our method with that of the others, we propose that the super-resolution method improves the sharpness and the clarity of WorldView-3 and KOMPSAT-2 images.

• Journal of the Korean Institute of Gas
• /
• v.16 no.6
• /
• pp.55-66
• /
• 2012

The practical way to minimize the greenhouse gas is to reduce the emission of carbon dioxide. For this reason, CCS(Carbon Capture and Storage) technology, which could reduce carbon dioxide emission, has risen as a realistic alternative in recent years. In addition, the researcher is recently working into ways of applying CCS technologies with deep saline aquifer. In this study, the evaluation model on the feasibility of $CO_2$ sequestration in the deep saline aquifer using ANN(Artificial Neural Network) was developed. In order to develop the efficiency-evaluation model, basic model was created in the deep saline aquifer and sensitivity analysis was performed for the aquifer characteristics by utilizing the commercial simulator of GEM. Based on the sensitivity analysis, the factors and ranges affecting $CO_2$ sequestration in the deep saline aquifer were chosen. The result from ANN training scenario were confirmed $CO_2$ sequestration by solubility trapping and residual trapping mechanism. The result from ANN model evaluation indicated there is the increase of correlation coefficient up to 0.99. It has been confirmed that the developed model can be utilized in feasibility of $CO_2$ sequestration at deep saline aquifer.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.14 no.6
• /
• pp.1249-1256
• /
• 2019

In this paper, we propose a new algorithm to remove acoustic feedback and noise in hearing aids. Instead of using the conventional FIR structure, this algorithm is a deep learning algorithm using neural network adaptive prediction filter to improve the feedback and noise reduction performance. The feedback canceller first removes the feedback signal from the microphone signal and then removes the noise using the Wiener filter technique. Noise elimination is to estimate the speech from the speech signal containing noise using the linear prediction model according to the periodicity of the speech signal. In order to ensure stable convergence of two adaptive systems in a loop, coefficient updates of the feedback canceller and noise canceller are separated and converged using the residual error signal generated after the cancellation. In order to verify the performance of the feedback and noise canceller proposed in this study, a simulation program was written and simulated. Experimental results show that the proposed deep learning algorithm improves the signal to feedback ratio(: SFR) of about 10 dB in the feedback canceller and the signal to noise ratio enhancement(: SNRE) of about 3 dB in the noise canceller than the conventional FIR structure.