• Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.37 no.3
• /
• pp.199-208
• /
• 2019

As the number of available satellites increases and technology advances, image information outputs are becoming increasingly diverse and a large amount of data is accumulating. In this study, we propose a change detection method for high-resolution satellite images that uses transfer learning and a deep learning network to overcome the limit caused by insufficient training data via the use of pre-trained information. The deep learning network used in this study comprises convolutional layers to extract the spatial and spectral information and convolutional long-short term memory layers to analyze the time series information. To use the learned information, the two initial convolutional layers of the change detection network are designed to use learned values from 40,000 patches of the ISPRS (International Society for Photogrammertry and Remote Sensing) dataset as initial values. In addition, 2D (2-Dimensional) and 3D (3-dimensional) kernels were used to find the optimized structure for the high-resolution satellite images. The experimental results for the KOMPSAT-3A (KOrean Multi-Purpose SATllite-3A) satellite images show that this change detection method can effectively extract changed/unchanged pixels but is less sensitive to changes due to shadow and relief displacements. In addition, the change detection accuracy of two sites was improved by using 3D kernels. This is because a 3D kernel can consider not only the spatial information but also the spectral information. This study indicates that we can effectively detect changes in high-resolution satellite images using the constructed image information and deep learning network. In future work, a pre-trained change detection network will be applied to newly obtained images to extend the scope of the application.

• Geophysics and Geophysical Exploration
• /
• v.5 no.3
• /
• pp.185-192
• /
• 2002

We made a study of 3-D migration which could precisely image data of GPR (Ground Penetrating Radar) applied to NDT (Non-Destructive Test) field for the inspection of structural safety. In this study, we obtained 3-D migrated images of important targets in structuresurvey (e.g. steel pipes, cracks) by using 3-D Kirchhoff prestack depth migration scheme developed for seismic data processing. For a concrete model consisting of steel pipe and void, the targets have been well defined with opposite amplitude according to the parameters of the targets. And migrated images using Parallel-Broadside array (XX configuration) have shown higher resolution than those using Perpendicular-Broadside array (YY configuration) when steel pipes had different sizes. Therefore, it is required to analyze the migrated image of XX configuration as well as that of general YY configuration in order to get more accurate information. As the last stage, we chose a model including two steel pipes which cross each other. The upper pipe has been resolved clearly but the lower has been imaged bigger than the model size due to the high conductivity of the upper steel.

• Journal of KIISE:Software and Applications
• /
• v.26 no.6
• /
• pp.791-803
• /
• 1999

본 논문은 두 장의 스테레요 영상으로부터 자동적으로 특징점 정합을 수행하도록 하는 새로운 절차의 효율적인 정합방법을 제안한다. 이를 위해 초기정합의 결과로 얻을 수 있는 유일 정합쌍을 이용한다. 즉, 본 논문에서는 초기정합의 결과로 얻어낸 유일 정합쌍의 정보를 이용하여 바로 outlier들을 제거시키므로써 초기정합의 결과가 갖는 애매성까지도 동반하여 상당량을 줄이도록 한다. 결국 애매성 제거에 대한 부담이 줄어들게 되므로 애매성 제거과정에서는 이완화 방법을 사용하지 않고 빠르게 애매성을 제거시킨다. 아울러 정합의 정확도를 높이기 위해 초기정합 후 바로 서브픽셀 정확도의 정합을 수행하며 정합의 마지막 단계에서는 추가정합을 수행하므로써 정합의 성능을 향상시킨다. 실내, 실회 스테레요 영상에 대한 다양한 실험결과는 본 논문에서 제안하는 방법의 특징점 정합기법이 빠르고 효율적임을 보여준다.

• Journal of the Korean Geophysical Society
• /
• v.4 no.3
• /
• pp.175-182
• /
• 2001

We consider the problem of determining smoothing parameters of Gibbs priors for Bayesian methods used in the medical imaging application of emission tomographic reconstruction. We address a simple smoothing prior (membrane) whose global hyperparameter (the smoothing parameter) controls the bias/variance tradeoff of the solution. We base our maximum-likelihood (ML) estimates of hyperparameters on observed training data, and argue the motivation for this approach. Good results are obtained with a simple ML estimate of the smoothing parameter for the membrane prior.

• The Journal of Engineering Research
• /
• v.4 no.1
• /
• pp.47-54
• /
• 2002

We consider the problem of determining smoothing parameters of Gibbs priors for Bayesian methods used in the medical imaging application of emission tomographic reconstruction. We address a simple smoothing prior (membrane) whose global hyperparameter (the smoothing parameter) controls the bias/variance tradeoff of the solution. We base our maximum-likelihood(ML) estimates of hyperparameters on observed training data, and argue the motivation for this approach. Good results are obtained with a simple ML estimate of the smoothing parameter for the membrane prior.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2020.07a
• /
• pp.656-659
• /
• 2020

초해상화 딥러닝 기법은 학습 시 수렴하기까지 최소 수백 번의 에폭을 필요로 하며 오랜 시간이 걸린다. 최근, 영상 인식용 딥러닝 모델에서는 학습 수렴 속도를 향상시키기 위해 픽셀, 채널간 불필요한 상호연관 정보를 제거하는 Deconvolution 기술이 제안되었다. 본 논문에서는 최초로 Deconvolution 기술을 초해상화 딥러닝 방법에 적용하여 학습 수렴 속도 증가를 시도했다. 영상 인식 딥러닝 기법과 다르게 초해상화 딥러닝 기법은 이미지 특성 추출 부분과 이미지 복원 부분의 정보를 보존하는 것이 중요하기 때문에, EDSR을 Baseline 모델로 사용하여 양쪽 끝의 레이어는 기존의 Convolution 연산을 그대로 유지하고, 중간 레이어의 ResBlock 내의 Convolution 연산만 Deconvolution 연산으로 바꿔서 구성하였다. 초해상화 벤치마크 데이터셋을 사용한 실험 결과, 수렴속도가 빨라지지 않는 결과를 도출했다. 본 논문에서는 Deconvolution 기술이 Baseline 모델의 성능을 개선하지 못하는 이유를 초해상화 분야에서 기본적으로 적용되는 Residual Learning 기법 때문으로 분석했다.

• Proceedings of the Korean Association of Geographic Inforamtion Studies Conference
• /
• 2003.04a
• /
• pp.334-339
• /
• 2003

위성 영상의 활용범위가 확대되면서 다양한 위성 센서로부터 위성영상이 제공되고 있다. 특히 최근에는 이기종 센서로부터 서로 다른 시간과 분광정보를 가진 영상의 자동 등록이 영상자료 분석을 위해 필요한 기술로 인식되고 있다. 본 연구에서는 Kompsat 영상과 Radarsat 영상을 이용하여 두 영상에서 공통으로 존재하는 패치(Patch)를 추출하고 그 패치의 중심점을 찾아 매칭하는 방법에 기초를 둔 자동영상 등록 기법을 제시하였다. 밝기 값분석을 통해 패치를 추출하고 추출된 패치를 모폴로지(Morphology)기법과 잡음요소 제거 기법을 적용하여 패치에 포함된 잡음을 제거하였으며, 비용함수를 이용한 패치 매칭과 변환함수를 이용하여 자동영상등록을 실시하였다.

• The Proceeding of the Korean Institute of Electromagnetic Engineering and Science
• /
• v.25 no.2
• /
• pp.3-9
• /
• 2014

• The Journal of Korean Institute of Electromagnetic Engineering and Science
• /
• v.21 no.11
• /
• pp.1285-1294
• /
• 2010

This study describes the design and verification of short range UWB(Ultra Wideband) imaging radar that is able to display high resolution radar image for front area of a UGV(Unmanned Ground Vehicle). This radar can help a UGV to navigate autonomously as it detects and avoids obstacles through foliage. We describe the relationship between bandwidth of transmitting signal and range resolution. A vivaldi antenna is designed and it's radiation pattern and reflection are measured. It is easy to make array antenna because of small size and thin shape. Aperture size of receiving array antenna is determined by azimuth resolution of radar image. The relation of interval of receiving antenna array, image resolution and aliasing of target on a radar image is analyzed. A vector network analyzer is used to obtain the reflected signal and corner reflectors as targets are positioned at grass field. Applicability of the proposed radar to UGV is proved by analysis of image resolution and penetrating capability for grass in the experiment.

• Korean Journal of Remote Sensing
• /
• v.30 no.2
• /
• pp.293-302
• /
• 2014

Radar systems are used in remote sensing mainly as space-borne, airborne and ground-based Synthetic Aperture Radar (SAR), scatterometer and Doppler radar. Those systems are composed of expensive equipments and require expertise and professional skills for operation. Because of the limitation in getting experiences of the radar and SAR systems and its operations in ordinary universities and institutions, it is difficult to learn and exercise essential principles of radar hardware which are essential to understand and develop new application fields. To overcome those difficulties, in this paper, we present the construction and experiment of a low-cost educational radar system based on the blueprints of the MIT Cantenna system. The radar system was operated in three modes. Firstly, the velocity of moving cars was measured in Doppler radar mode. Secondly, the range of two moving targets were measured in radar mode with range resolution. Lastly, 2D images were constructed in GB-SAR mode to enhance the azimuth resolution. Additionally, we simulated the SAR raw data to compare Deramp-FFT and ${\omega}-k$ algorithms and to analyze the effect of antenna positional error for SAR focusing. We expect the system can be further developed into a light-weight SAR system onboard a unmanned aerial vehicle by improving the system with higher sampling frequency, I/Q acquisition, and more stable circuit design.