• Journal of the Korea Institute of Military Science and Technology
• /
• v.22 no.2
• /
• pp.162-169
• /
• 2019

This paper describes a thermal image processing algorithm for uncooled type TEC-less IR detector which is applicable to fire control system of small arms. We implemented a real-time gain and offset compensation algorithm based on polynomial approximation from the raw dataset which is acquired by two reference temperature of blackbody from various FPA(Focal Plane Array) temperature. Through the experiment, we analyzed the output characteristics of detector's raw-data and compared IR image quality to traditional non-uniformity correction method. It shows that the proposed method works well in all FPA temperature range with low residual non-uniformity.

• International Journal of Computer Science & Network Security
• /
• v.22 no.1
• /
• pp.1-6
• /
• 2022

Exploration has always been an instinct of humans, and underwater life is as fascinating as it seems. So, for studying flora and fauna below water, there is a need for high-quality images. However, the underwater images tend to be of impaired quality due to various factors, which calls for improved and enhanced underwater images. There are various Histogram Equalization (HE) based techniques which could aid in solving these issues. Classifying the HE methods broadly, there is Global Histogram Equalization (GHE), Mean Brightness Preserving HE (MBPHE), Bin Modified HE (BMHE), and Local HE (LHE). Each of these HE extensions have their own pros and cons and thus, by considering them we have considered BBHE, CLAHE, BPDHE, BPDFHE, and DSIHE enhancement algorithms, which are based on Mean Brightness Preserving HE and Local HE, for this study. The performance is evaluated with non-reference performance measures like Entropy, UCIQE, UICM, and UIQM. In this study, we apply the enhancement algorithms on 300 images from the UIEB benchmark dataset and then apply the techniques of cascading fusion on the best-performing algorithms.

• IEMEK Journal of Embedded Systems and Applications
• /
• v.18 no.1
• /
• pp.19-24
• /
• 2023

In traditional fish farming way, the workers have to observe all of the pools every time and every day to feed at the right timing. This method causes tremendous stress on workers and wastes time. To solve this problem, we implemented an automatic detection system for feeding time using deep neural network. The detection system consists of two steps: classification of the presence or absence of feed and checking DO (Dissolved Oxygen) of the pool. For the classification, the pretrained ResNet18 model and transfer learning with custom dataset are used. DO is obtained from the DO sensor in the pool through HTTP in real time. For better accuracy, the next step, checking DO proceeds when the result of the classification is absence of feed several times in a row. DO is checked if it is higher than a DO reference value that is set by the workers. These actions are performed automatically in the UI programs developed with LabVIEW.

• Nuclear Engineering and Technology
• /
• v.54 no.12
• /
• pp.4751-4758
• /
• 2022

Two-phase flow may almost exist in every branch of the energy industry. For the corresponding engineering design, it is very essential and crucial to monitor flow patterns and their transitions accurately. With the high-speed development and success of deep learning based on convolutional neural network (CNN), the study of flow pattern identification recently almost focused on this methodology. Additionally, the photographing technique has attractive implementation features as well, since it is normally considerably less expensive than other techniques. The development of such a two-phase flow pattern online monitoring system is the objective of this work, which seldom studied before. The ongoing preliminary engineering design (including hardware and software) of the system are introduced. The flow pattern identification method based on CNNs and transfer learning was discussed in detail. Several potential CNN candidates such as ALexNet, VggNet16 and ResNets were introduced and compared with each other based on a flow pattern dataset. According to the results, ResNet50 is the most promising CNN network for the system owing to its high precision, fast classification and strong robustness. This work can be a reference for the online monitoring system design in the energy system.

• The Journal of Korea Robotics Society
• /
• v.18 no.2
• /
• pp.143-154
• /
• 2023

In this study, a fast motion planning method for the swing motion of a 6x6 wheel-legged robot to traverse large obstacles and gaps is proposed. The motion planning method presented in the previous paper, which was based on trajectory optimization, took up to tens of seconds and was limited to two-dimensional, structured vertical obstacles and trenches. A deep neural network based on one-dimensional Convolutional Neural Network (CNN) is introduced to generate keyframes, which are then used to represent smooth reference commands for the six leg angles along the robot's path. The network is initially trained using the behavioral cloning method with a dataset gathered from previous simulation results of the trajectory optimization. Its performance is then improved through reinforcement learning, using a one-step REINFORCE algorithm. The trained model has increased the speed of motion planning by up to 820 times and improved the success rates of obstacle crossing under harsh conditions, such as low friction and high roughness.

• Computers and Concrete
• /
• v.32 no.4
• /
• pp.399-410
• /
• 2023

This study proposes a novel use of backpropagated Levenberg-Marquardt neural networks based on computational intelligence heuristics to comprehend the examination of hybrid nanoparticles on the flow of dusty liquid via stretched cylinder. A two-phase model is employed in the present work to describe the fluid flow. The use of desulphated nanoparticles of silver and molybdenum suspended in water as base fluid. The mathematical model represented in terms of partial differential equations, Implementing similarity transformationsis model is converted to ordinary differential equations for the analysis . By adjusting the particle mass concentration and curvature parameter, a unique technique is utilized to generate a dataset for the proposed Levenberg-Marquardt neural networks in various nanoparticle circumstances on the flow of dusty liquid via stretched cylinder. The intelligent solver Levenberg-Marquardt neural networks is trained, tested and verified to identify the nanoparticles on the flow of dusty liquid solution for various situations. The Levenberg-Marquardt neural networks approach is applied for the solution of the hybrid nanoparticles on the flow of dusty liquid via stretched cylinder model. It is validated by comparison with the standard solution, regression analysis, histograms, and absolute error analysis. Strong agreement between proposed results and reference solutions as well as accuracy provide an evidence of the framework's validity.

• Korean Journal of Remote Sensing
• /
• v.29 no.2
• /
• pp.161-172
• /
• 2013

The construction of hyperspectral test-bed dataset is essential for the effective performance of hyperspectral image for various applications. In this study, we analyzed the technical points for generating of optimal hyperspectral test-bed site for hyperspectral sensors and the efficiency of hyperspectral test-bed site. In this regard regions we analyzed existing construction techniques for generating test-bed site in domestic and foreign, and designed the test-bed site to acquire images from the airborne hyperspectral sensor. To produce a reference data from the image of constructed test-bed site, this study applied vicarious correction as a pre-processing and analyzed its efficiency. The result presented that it was ideal to use tarp for the vicarious correction, but it is possible to use the materials with constant spectral reflectance or with relatively low variance of spectral reflectance. The test-bed data taken in this study can be employed as the reference of domestic and foreign studies for hyperspectral image processing.

• Korean Journal of Radiology
• /
• v.23 no.12
• /
• pp.1251-1259
• /
• 2022

Objective: T1 mapping provides valuable information regarding cardiomyopathies. Manual drawing is time consuming and prone to subjective errors. Therefore, this study aimed to test a DL algorithm for the automated measurement of native T1 and extracellular volume (ECV) fractions in cardiac magnetic resonance (CMR) imaging with a temporally separated dataset. Materials and Methods: CMR images obtained for 95 participants (mean age ± standard deviation, 54.5 ± 15.2 years), including 36 left ventricular hypertrophy (12 hypertrophic cardiomyopathy, 12 Fabry disease, and 12 amyloidosis), 32 dilated cardiomyopathy, and 27 healthy volunteers, were included. A commercial deep learning (DL) algorithm based on 2D U-net (Myomics-T1 software, version 1.0.0) was used for the automated analysis of T1 maps. Four radiologists, as study readers, performed manual analysis. The reference standard was the consensus result of the manual analysis by two additional expert readers. The segmentation performance of the DL algorithm and the correlation and agreement between the automated measurement and the reference standard were assessed. Interobserver agreement among the four radiologists was analyzed. Results: DL successfully segmented the myocardium in 99.3% of slices in the native T1 map and 89.8% of slices in the post-T1 map with Dice similarity coefficients of 0.86 ± 0.05 and 0.74 ± 0.17, respectively. Native T1 and ECV showed strong correlation and agreement between DL and the reference: for T1, r = 0.967 (95% confidence interval [CI], 0.951-0.978) and bias of 9.5 msec (95% limits of agreement [LOA], -23.6-42.6 msec); for ECV, r = 0.987 (95% CI, 0.980-0.991) and bias of 0.7% (95% LOA, -2.8%-4.2%) on per-subject basis. Agreements between DL and each of the four radiologists were excellent (intraclass correlation coefficient [ICC] of 0.98-0.99 for both native T1 and ECV), comparable to the pairwise agreement between the radiologists (ICC of 0.97-1.00 and 0.99-1.00 for native T1 and ECV, respectively). Conclusion: The DL algorithm allowed automated T1 and ECV measurements comparable to those of radiologists.

• Korean Journal of Remote Sensing
• /
• v.38 no.6_1
• /
• pp.1057-1068
• /
• 2022

This paper compares the combination performance of feature point-based matching algorithms as a study to confirm the matching possibility between image taken by a user and a virtual texture image with the goal of developing mobile-based real-time image positioning technology. The feature based matching algorithm includes process of extracting features, calculating descriptors, matching features from both images, and finally eliminating mismatched features. At this time, for matching algorithm combination, we combined the process of extracting features and the process of calculating descriptors in the same or different matching algorithm respectively. V-World 3D desktop was used for the virtual indoor texture image. Currently, V-World 3D desktop is reinforced with details such as vertical and horizontal protrusions and dents. In addition, levels with real image textures. Using this, we constructed dataset with virtual indoor texture data as a reference image, and real image shooting at the same location as a target image. After constructing dataset, matching success rate and matching processing time were measured, and based on this, matching algorithm combination was determined for matching real image with virtual image. In this study, based on the characteristics of each matching technique, the matching algorithm was combined and applied to the constructed dataset to confirm the applicability, and performance comparison was also performed when the rotation was additionally considered. As a result of study, it was confirmed that the combination of Scale Invariant Feature Transform (SIFT)'s feature and descriptor detection had the highest matching success rate, but matching processing time was longest. And in the case of Features from Accelerated Segment Test (FAST)'s feature detector and Oriented FAST and Rotated BRIEF (ORB)'s descriptor calculation, the matching success rate was similar to that of SIFT-SIFT combination, while matching processing time was short. Furthermore, in case of FAST-ORB, it was confirmed that the matching performance was superior even when 10° rotation was applied to the dataset. Therefore, it was confirmed that the matching algorithm of FAST-ORB combination could be suitable for matching between virtual texture image and real image.

• Korean Journal of Remote Sensing
• /
• v.37 no.2
• /
• pp.221-232
• /
• 2021

This study analyzed co-registration results according to the geometric processing level of reference image, which are Level 1R and Level 1G provided from KOMPSAT-3 and KOMPSAT-3A images. We performed co-registration using each Level 1R and Level 1G image as a reference image, and Level 1R image as a sensed image. For constructing the experimental dataset, seven Level 1R and 1G images of KOMPSAT-3 and KOMPSAT-3A acquired from Daejeon, South Korea, were used. To coarsely align the geometric position of the two images, SURF (Speeded-Up Robust Feature) and PC (Phase Correlation) methods were combined and then repeatedly applied to the overlapping region of the images. Then, we extracted tie-points using the SURF method from coarsely aligned images and performed fine co-registration through affine transformation and piecewise Linear transformation, respectively, constructed with the tie-points. As a result of the experiment, when Level 1G image was used as a reference image, a relatively large number of tie-points were extracted than Level 1R image. Also, in the case where the reference image is Level 1G image, the root mean square error of co-registration was 5 pixels less than the case of Level 1R image on average. We have shown from the experimental results that the co-registration performance can be affected by the geometric processing level related to the initial geometric relationship between the two images. Moreover, we confirmed that the better geometric quality of the reference image achieved the more stable co-registration performance.