• Smart Structures and Systems
• /
• v.29 no.1
• /
• pp.1-16
• /
• 2022

Despite recent breakthroughs in deep learning and computer vision fields, the pixel-wise identification of tiny objects in high-resolution images with complex disturbances remains challenging. This study proposes a modified U-net for tiny crack segmentation in real-world steel-box-girder bridges. The modified U-net adopts the common U-net framework and a novel Self-Attention-Self-Adaption (SASA) neuron as the fundamental computing element. The Self-Attention module applies softmax and gate operations to obtain the attention vector. It enables the neuron to focus on the most significant receptive fields when processing large-scale feature maps. The Self-Adaption module consists of a multiplayer perceptron subnet and achieves deeper feature extraction inside a single neuron. For data augmentation, a grid-based crack random elastic deformation (CRED) algorithm is designed to enrich the diversities and irregular shapes of distributed cracks. Grid-based uniform control nodes are first set on both input images and binary labels, random offsets are then employed on these control nodes, and bilinear interpolation is performed for the rest pixels. The proposed SASA neuron and CRED algorithm are simultaneously deployed to train the modified U-net. 200 raw images with a high resolution of 4928 × 3264 are collected, 160 for training and the rest 40 for the test. 512 × 512 patches are generated from the original images by a sliding window with an overlap of 256 as inputs. Results show that the average IoU between the recognized and ground-truth cracks reaches 0.409, which is 29.8% higher than the regular U-net. A five-fold cross-validation study is performed to verify that the proposed method is robust to different training and test images. Ablation experiments further demonstrate the effectiveness of the proposed SASA neuron and CRED algorithm. Promotions of the average IoU individually utilizing the SASA and CRED module add up to the final promotion of the full model, indicating that the SASA and CRED modules contribute to the different stages of model and data in the training process.

• Journal of the Korean Society of Radiology
• /
• v.17 no.3
• /
• pp.335-341
• /
• 2023

In this study, four algorithms (Standard, Bone, Dual volume, and Stent Follow up) were applied to the image of the aneurysm in cerebral angiography to reconstruct the image in 3D, and quantitatively evaluate Noise, SNR, and CNR based on the reconstructed image to find out the optimal algorithm. As an analysis method, Image J program, which can analyze images and calculate area and pixel values, was used for images reconstructed with four algorithms. In order to obtain Noise, SNR, and CNR, the region of interest (ROI) is measured by designating the point where the abnormal artery (aneurysm) is located and the surrounding normal artery in the image are measured, and the mean value and SD value are obtained. Background noise was set to two surrounding normal artery to increase reliability. The values of SNR and CNR were calculated based on the given formula. As a result, the noise was the lowest in the stent follow-up algorithm, and the SNR and CNR were the highest. Therefore, the stent follow-up algorithm is judged to be the most appropriate algorithm. The data of this study are expected to be useful as basic data for 3D image evaluation of the vascular and aneurysm in cerebral angiography, and it is believed that appropriate algorithm changes will serve as an opportunity to further improve image quality.

• KSCE Journal of Civil and Environmental Engineering Research
• /
• v.26 no.4D
• /
• pp.687-693
• /
• 2006

Image co-registration is the process of overlaying two images of the same scene, one of which represents a reference image, while the other is geometrically transformed to the one. In order to improve efficiency and effectiveness of the co-registration approach, the author proposed a pre-qualified area matching algorithm which is composed of feature extraction with canny operator and area matching algorithm with cross correlation coefficient. For refining matching points, outlier detection using studentized residual was used and iteratively removes outliers at the level of three standard deviation. Throughout the pre-qualification and the refining processes, the computation time was significantly improved and the registration accuracy is enhanced. A prototype of the proposed algorithm was implemented and the performance test of 3 Landsat images of Korea. showed: (1) average RMSE error of the approach was 0.435 pixel; (2) the average number of matching points was over 25,573; (3) the average processing time was 4.2 min per image with a regular workstation equipped with a 3 GHz Intel Pentium 4 CPU and 1 Gbytes Ram. The proposed approach achieved robustness, full automation, and time efficiency.

• Journal of the Korean Society of Radiology
• /
• v.17 no.5
• /
• pp.679-684
• /
• 2023

Venous puncture is widely used to obtain blood samples for pathological examination. Because the invasive venipuncture method using a needle is repeatedly performed, the pain suffered by the patient increases, so our research team pre-developed a miniaturized near-infrared (NIR) imaging system in advance. To improve the image quality of the acquired NIR images, this study aims to model the non-local means (NLM) algorithm, which is well known to be efficient in noise reduction, and analyze its applicability in the system. The developed NIR imaging system is based on the principle that infrared rays pass through dichroic and long-pass filters and are detected by a CMOS sensor module. The proposed NLM algorithm is modeled based on the principle of replacing the pixel from which noise is to be removed with a value that reflects the distances between surrounding pixels. After acquiring an NIR image with a central wavelength of 850 nm, the NLM algorithm was applied to segment the final vein area through histogram equalization. As a result, the coefficient of variation of the NIR image of the vein using the NLM algorithm was 0.247 on average, which was an excellent result compared to conventional filtering methods. In addition, the dice similarity coefficient value of the NLM algorithm was improved by 62.91 and 9.40%, respectively, compared to the median filter and total variation methods. In conclusion, we demonstrated that the NLM algorithm can acquire accurate segmentation of veins acquired with a NIR imaging system.

• Journal of Korean Tunnelling and Underground Space Association
• /
• v.26 no.4
• /
• pp.365-384
• /
• 2024

The development of scanning technology is accelerating for safer and more efficient automated inspection than human-based inspection. Research on automatically detecting facility damage from images collected using computer vision technology is also increasing. The pixel size, quality, and quantity of an image can affect the performance of deep learning or image processing for automatic damage detection. This study is a basic to acquire high-quality raw image data and camera performance of a mobile tunnel scanning system for automatic detection of damage based on deep learning, and proposes a method to quantitatively evaluate image quality. A test chart was attached to a panel device capable of simulating a moving speed of 40 km/h, and an indoor test was performed using the international standard ISO 12233 method. Existing image quality evaluation methods were applied to evaluate the quality of images obtained in indoor experiments. It was determined that the shutter speed of the camera is closely related to the motion blur that occurs in the image. Modulation transfer function (MTF), one of the image quality evaluation method, can objectively evaluate image quality and was judged to be consistent with visual observation.

• Journal of the Korean Society of Earth Science Education
• /
• v.17 no.2
• /
• pp.168-180
• /
• 2024

In this study, we developed a STEAM program based on digital astronomical and space resources and applied it to 167 randomly selected elementary school students to examine the effectiveness of the STEAM program in enhancing creative problem solving skills. The results of this study are as follows: first, the statistical analysis of pre-test and post-test scores for creative problem-solving skills related to the STEAM program showed significant results (p<.05). This suggests that the STEAM program contributed to improving creative problem-solving skills through the procedural process involved in generating ideas and convergent thinking during the digital problem-solving. Second, the paired sample t-test based on the pre-test and post-test of the STEAM attitude test also showed significant results (p<.05). Analyzing digital materials and presenting pixel art projects positively influenced STEAM attitudes in terms of interest, communication, and usefulness. This underscores the need for developing integrative education programs utilizing advanced technologies in the future. Third, in the class satisfaction test conducted after the application of the STEAM program, the satisfaction factor scored an average of 3.55, interest scored 3.35, and overall class difficulty scored 3.46. The main difference from traditional classes was the focus on 'acquisition of future career information.' However, given the slight decrease in interest during the digital transformation and interpretation process, it is recommended that future classes allocate sufficient time for experiential activities. To generalize the developed program, future studies should consider various school levels, implementation periods, and difficulty levels.

• Progress in Medical Physics
• /
• v.20 no.2
• /
• pp.72-79
• /
• 2009

Scatter correction for I-131 plays a very important role to improve image quality and quantitation. I-131 has multiple and higher energy gamma-ray emissions. Image quality and quantitative accuracy in I-131 imaging are degraded by object scatter as well as scatter and septal penetration in the collimator. The purpose of this study was to estimate scatter and septal penetration and investigate two scatter correction methods using Monte Carlo simulation. The gamma camera system simulated in this study was a FORTE system (Phillips, Nederland) with high energy, general-purpose, parallel hole collimator. We simulated for two types of high energy collimators. One is composed of lead, and the other is composed of artificially high Z number and high density. We simulated energy spectrum using a point source in air. We estimated both full width at half maximum (FWHM) and full width at tenth maximum (FWTM) using line spread function (LSF) in cylindrical water phantom. We applied two scatter correction methods, triple energy window scatter correction (TEW) and extended triple energy window scatter correction (ETEW). The TEW method is a pixel-by pixel based correction which is easy to implement clinically. The ETEW is a modification of the TEW which corrects for scatter by using abutted scatter rejection window, which can overestimate or the underestimate scatter. The both FWHM and FWTM were estimated as 41.2 mm and 206.5 mm for lead collimator, respectively. The FWHM and FWTM were estimated as 27.3 mm and 45.6 mm for artificially high Z and high density collimator, respectively. ETEW showed that the estimation of scatter components was close to the true scatter components. In conclusion, correction for septal penetration and scatter is important to improve image quality and quantitative accuracy in I-131 imaging. The ETEW method in scatter correction appeared to be useful in I-131 imaging.

• Korean Journal of Remote Sensing
• /
• v.39 no.6_1
• /
• pp.1283-1297
• /
• 2023

Mid-wave infrared (MWIR) imagery, due to its ability to capture the temperature of land cover and objects, serves as a crucial data source in various fields including environmental monitoring and defense. The KOMPSAT-3A satellite acquires MWIR imagery with high spatial resolution compared to other satellites. However, the limited spatial resolution of MWIR imagery, in comparison to electro-optical (EO) imagery, constrains the optimal utilization of the KOMPSAT-3A data. This study aims to create a highly visible MWIR fusion image by leveraging the edge information from the KOMPSAT-3A panchromatic (PAN) image. Preprocessing is implemented to mitigate the relative geometric errors between the PAN and MWIR images. Subsequently, we employ a pre-trained pixel difference network (PiDiNet), a deep learning-based edge information extraction technique, to extract the boundaries of objects from the preprocessed PAN images. The MWIR fusion imagery is then generated by emphasizing the brightness value corresponding to the edge information of the PAN image. To evaluate the proposed method, the MWIR fusion images were generated in three different sites. As a result, the boundaries of terrain and objects in the MWIR fusion images were emphasized to provide detailed thermal information of the interest area. Especially, the MWIR fusion image provided the thermal information of objects such as airplanes and ships which are hard to detect in the original MWIR images. This study demonstrated that the proposed method could generate a single image that combines visible details from an EO image and thermal information from an MWIR image, which contributes to increasing the usage of MWIR imagery.

• Progress in Medical Physics
• /
• v.24 no.3
• /
• pp.162-170
• /
• 2013

Dedicated single-photon emission computed tomography (SPECT) systems based on pixelated semiconductors are being developed for studying small animal models of human disease. To clarify the possibility of using a SPECT system with CdTe for a high resolution low-dose small animal imaging, we compared the quality of reconstructed images from pixelated CdTe detector to those from a small SPECT system with NaI(Tl). The CdTe detector was $44.8{\times}44.8$ mm and the pixels were $0.35{\times}0.35{\times}5$ mm. The intrinsic resolution of the detector was 0.35 mm, which is equal to the pixel size. GATE simulations were performed to assess the image quality of both SPECT systems. The spatial resolutions and sensitivities for both systems were evaluated using a 10 MBq $^{99m}Tc$ point source. The quantitative comparison with different injected dose was performed using a voxelized MOBY phantom, and the absorbed doses for each organ were evaluated. The spatial resolution of the SPECT with NaI(Tl) was about 1.54 mm FWHM, while that of the SPECT with a CdTe detector was about 1.32 mm FWHM at 30 mm. The sensitivity of NaI(Tl) based SPECT was 83 cps/MBq, while that of the CdTe detector based SPECT was 116 cps/MBq at 30 mm. The image statistics were evaluated by calculating the CNR of the image from both systems. When the injected activity for the striatum in the mouse brain was 160 Bq/voxel, the CNR of CdTe based SPECT was 2.30 while that of NaI(Tl) based SPECT was 1.85. The CNR of SPECT with CdTe was overall higher than that of the NaI(Tl) based SPECT. In addition, the absorbed dose was higher from SPECT with CdTe than those from NaI(Tl) based SPECT to acquire the same quantitative values. Our simulation results indicated that the SPECT with CdTe detector showed overall high performance compared to the SPECT with NaI(Tl). Even though the validation study is needed, the SPECT system with CdTe detector appeared to be feasible for high resolution low-dose small animal imaging.

• Korean Journal of Remote Sensing
• /
• v.34 no.2_2
• /
• pp.377-391
• /
• 2018

In order to efficiently monitor red tide over a wide range, the need for red tide detection using remote sensing is increasing. However, the previous studies focus on the development of red tide detection algorithm for ocean colour sensor. In this study, we propose the use of multi-sensor to improve the inaccuracy for red tide detection and remote sensing data in coastal areas with high turbidity, which are pointed out as limitations of satellite-based red tide monitoring. The study area were selected based on the red tide information provided by National Institute of Fisheries Science, and spatial fusion and spectral-based fusion were attempted using GOCI image as ocean colour sensor and Landsat OLI image as terrestrial sensor. Through spatial fusion of the two images, both the red tide of the coastal area and the outer sea areas, where the quality of Landsat OLI image was low, which were impossible to observe in GOCI images, showed improved detection results. As a result of spectral-based fusion performed by feature-level and rawdata-level, there was no significant difference in red tide distribution patterns derived from the two methods. However, in the feature-level method, the red tide area tends to overestimated as spatial resolution of the image low. As a result of pixel segmentation by linear spectral unmixing method, the difference in the red tide area was found to increase as the number of pixels with low red tide ratio increased. For rawdata-level, Gram-Schmidt sharpening method estimated a somewhat larger area than PC spectral sharpening method, but no significant difference was observed. In this study, it is shown that coastal red tide with high turbidity as well as outer sea areas can be detected through spatial fusion of ocean colour and terrestrial sensor. Also, by presenting various spectral-based fusion methods, more accurate red tide area estimation method is suggested. It is expected that this result will provide more precise detection of red tide around the Korean peninsula and accurate red tide area information needed to determine countermeasure to effectively control red tide.