• Smart Structures and Systems
• /
• v.29 no.1
• /
• pp.153-166
• /
• 2022

Identifying fine cracks in steel bridge facilities is a challenging task of structural health monitoring (SHM). This study proposed an end-to-end crack image segmentation framework based on a one-step Convolutional Neural Network (CNN) for pixel-level object recognition with high accuracy. To particularly address the challenges arising from small object detection in complex background, efforts were made in loss function selection aiming at sample imbalance and module modification in order to improve the generalization ability on complicated images. Specifically, loss functions were compared among alternatives including the Binary Cross Entropy (BCE), Focal, Tversky and Dice loss, with the last three specialized for biased sample distribution. Structural modifications with dilated convolution, Spatial Pyramid Pooling (SPP) and Feature Pyramid Network (FPN) were also performed to form a new backbone termed CrackDet. Models of various loss functions and feature extraction modules were trained on crack images and tested on full-scale images collected on steel box girders. The CNN model incorporated the classic U-Net as its backbone, and Dice loss as its loss function achieved the highest mean Intersection-over-Union (mIoU) of 0.7571 on full-scale pictures. In contrast, the best performance on cropped crack images was achieved by integrating CrackDet with Dice loss at a mIoU of 0.7670.

• Korean Journal of Remote Sensing
• /
• v.37 no.5_1
• /
• pp.1135-1147
• /
• 2021

In order to geometrically correct high-resolution satellite imagery, the sensor modeling process that restores the geometric relationship between the satellite sensor and the ground surface at the image acquisition time is required. In general, high-resolution satellites provide RPC (Rational Polynomial Coefficient) information, but the vendor-provided RPC includes geometric distortion caused by the position and orientation of the satellite sensor. GCP (Ground Control Point) is generally used to correct the RPC errors. The representative method of acquiring GCP is field survey to obtain accurate ground coordinates. However, it is difficult to find the GCP in the satellite image due to the quality of the image, land cover change, relief displacement, etc. By using image maps acquired from various sensors as reference data, it is possible to automate the collection of GCP through the image matching algorithm. In this study, the RPC of KOMPSAT-3A satellite image was corrected through the extracted matching point using the UAV (Unmanned Aerial Vehichle) imagery. We propose a pre-porocessing method for the extraction of matching points between the UAV imagery and KOMPSAT-3A satellite image. To this end, the characteristics of matching points extracted by independently applying the SURF (Speeded-Up Robust Features) and the phase correlation, which are representative feature-based matching method and area-based matching method, respectively, were compared. The RPC adjustment parameters were calculated using the matching points extracted through each algorithm. In order to verify the performance and usability of the proposed method, it was compared with the GCP-based RPC correction result. The GCP-based method showed an improvement of correction accuracy by 2.14 pixels for the sample and 5.43 pixelsfor the line compared to the vendor-provided RPC. In the proposed method using SURF and phase correlation methods, the accuracy of sample was improved by 0.83 pixels and 1.49 pixels, and that of line wasimproved by 4.81 pixels and 5.19 pixels, respectively, compared to the vendor-provided RPC. Through the experimental results, the proposed method using the UAV imagery presented the possibility as an alternative to the GCP-based method for the RPC correction.

• Korean Journal of Remote Sensing
• /
• v.33 no.1
• /
• pp.15-24
• /
• 2017

Jellyfish (cnidarian) are conspicuous in many marine ecosystems when in bloom. Despite their importance for the ecosystem structure and function, very few sampling programs are dedicated to sample jellyfish because they are patchily distributed and easily clogged plankton net. Although satellite remote sensing is an excellent observing tool for many phenomena in the ocean, their uses for monitoring jellyfish are not possible due to the coarse spatial resolutions. Hence, we developed the low altitude remote sensing platform to detect jellyfish in high resolutions, which allow us to monitor not only horizontal, but also vertical migration of them. Using low altitude remote sensing platform,we measured the jellyfish from the pier at the Chesapeake Biological Laboratory in Chesapeake Bay. The patterns observed included discrete patches, in rows that were aligned with waves that propagated from deeper regions, and aggregation around physical objects. The corresponding areas of exposed jellyfish on the sea surface were $0.1{\times}10^4pixel^2$, $0.3{\times}10^4pixel^2$, and $2.75{\times}10^4pixel^2$, respectively. Thus, the research result suggested that the migration of the jellyfish was related to the physical forcing in the sea surface.

• Journal of the Institute of Electronics Engineers of Korea SD
• /
• v.44 no.3 s.357
• /
• pp.65-71
• /
• 2007

We propose a pixel-level automatic calibration circuit scheme that initializes a capacitive fingerprint sensor LSI to eliminate the influence of the surface condition and environment, which is degraded by dirt during long-time use, process variation and ambient temperature. The sample chip is fabricated on $0.35{\mu}m$ standard CMOS process. The calibration is executed by optimizing the reference voltage in each pixel to make the sensor signals of all pixels the same. The calibration control circuit is composed of the sensing circuit and charge pumping circuit, and calibrates all pixels in a short time. 16-level gray scale fingerprint images can be captured to increase the accuracy of identification. This confirms that the scheme is effective for capturing consistent clear images during long-time use.

• Nuclear Engineering and Technology
• /
• v.53 no.3
• /
• pp.888-893
• /
• 2021

The International Atomic Energy Agency has developed a tomographic imaging system for accomplishing the total fuel rod-by-rod verification time of fuel assemblies within the order of 1-2 h, however, there are still limitations for some fuel types. The aim of this study is to develop a deep learning-based denoising process resulting in increasing the tomographic image acquisition speed of fuel assembly compared to the conventional techniques. Convolutional AutoEncoder (CAE) was employed for denoising the low-quality images reconstructed by filtered back-projection (FBP) algorithm. The image data set was constructed by the Monte Carlo method with the FBP and ground truth (GT) images for 511 patterns of missing fuel rods. The de-noising performance of the CAE model was evaluated by comparing the pixel-by-pixel subtracted images between the GT and FBP images and the GT and CAE images; the average differences of the pixel values for the sample image 1, 2, and 3 were 7.7%, 28.0% and 44.7% for the FBP images, and 0.5%, 1.4% and 1.9% for the predicted image, respectively. Even for the FBP images not discriminable the source patterns, the CAE model could successfully estimate the patterns similarly with the GT image.

• The Korean Journal of Applied Statistics
• /
• v.14 no.1
• /
• pp.149-160
• /
• 2001

In this paper we employ two-sample location tests such as Wilcoxon test and T test for detecting edges in noisy images. For this, we compute a test statistic on pixel gray levels obtained using an edge-height parameter and compare it with a threshold determined by a significance level. Experimental results applied to sample images are given and performances of these tests in terms of the objective measure are compared.

• Journal of the Optical Society of Korea
• /
• v.1 no.2
• /
• pp.90-93
• /
• 1997

The shallow depth of focus in conventional light microscopy hinders the observation of the whole image when the object is thicker than the depth of field. Most of the existing techniques measured the object distance, which is not necessarily the actual distance of each pixel in the image. We implemented a means of determining the "best focus" of each pixel and located the height of object points by sectioning at different sample heights. Combining the height information and its gray values together, we obtained an image where the blur from the finite depth of focus is eliminated. Limitations of the technique are discussed together with composed images.ed images.

• 전기의세계
• /
• v.30 no.12
• /
• pp.793-796
• /
• 1981

최근 우리나라에도 digital image processing에 대한 연구가 활발이 진행되고 있다. 또한 digital image processing의 생체공학에의 응용도 괄목할 만하다. 이러한 연구와 응용에 도움이 되고자 microcomputer를 이용한 image processor설계의 실례를 기술하고자 한다. 이 설계는 목적 image를 stationary image로 가정하여 TV카메라의 영상신호를 sample하여 computer 기억장치에 저장하므로 가격이 저렴하게 된다. 이러한 장치로서는 DMA연결 (Direct Memory Access interface)을 사용하여 빠른 data transfer를 달성할 수 있다. Digital image processing계는 기본적으로 microcomputer가 TV카메라와 TV모니터에 연결된 구조를 하고 있다. Computer가 기억장치에 저장된 data를 처리하여 필요한 정보를 얻게 된다. 이러한 data 처리를 하므로서 image를 사용자가 해석하기 쉽도록 image질을 향상시키거나 computer가 image를 인식하게 한다. 이와같이 처리된 image는 TV모니터를 통해서 볼 수 있다. 본지에서는 256x256개의 pixel들로 이루어지고, 각개의 pixel은 4개의 bit로 구성된 image processor의 설계를 기술한다.

• The Transactions of the Korea Information Processing Society
• /
• v.7 no.2
• /
• pp.477-487
• /
• 2000

When volume data is visualized by the ray casting method, the color value of each pixel in the image is obtained by composing the color contributions of the sample points that lie on the ray cast from the pixel point. In most ray tracing methods including Levoy's classical method, the color composition is formulated as a summation of the color contributions of the discrete sample points. However, the more precise color composition is formulated as differential equations over the color contributions of the continuous sample points. The discrete formulation is used, because analytical solutions to the continuous formulations are hard to find. In this paper, however, we have discovered a semi-analytical solution to the continuous formulation of a typical ray tracing of volume data. We have applied both Levoy's method and ours to the same set of data, and compared the visual quality of both results. The comparison shows that our method produces a more fine-grained visualization of volume data.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.14 no.4
• /
• pp.1648-1672
• /
• 2020

There are several types of image registration in the sense of stitching separated images that overlap each other. One of these is feature-based registration by a common feature descriptor. In this study, we generate a mosaic of images using feature-based registration for drone aerial images. As a feature descriptor, we apply the scale-invariant feature transform descriptor. In order to investigate the authenticity of the feature points and to have the mapping function, we employ the sample consensus method; we consider the sensed image's inherent characteristic such as the geometric congruence between the feature points of the images to propose a novel hypothesis estimation of the mapping function of the stitching via some optimally chosen initial candidate inliers in the sample consensus method. Based on the experimental results, we show the efficiency of the proposed method compared with benchmark methodologies of random sampling consensus method (RANSAC); the well-ordering property defined in the context and the extensive stitching examples have supported the utility. Moreover, the sample consensus scheme proposed in this study is uncomplicated and robust, and some fatal miss stitching by RANSAC is remarkably reduced in the measure of the pixel difference.