• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.17 no.10
• /
• pp.2643-2657
• /
• 2023

Scene graphs are structured representations that can clearly convey objects and the relationships between them, but are often heavily biased due to the highly skewed, long-tailed relational labeling in the dataset. Indeed, the visual world itself and its descriptions are biased. Therefore, Unbiased Scene Graph Generation (USGG) prefers to train models to eliminate long-tail effects as much as possible, rather than altering the dataset directly. To this end, we propose Geometric and Semantic Improvement (GSI) for USGG to mitigate this issue. First, to fully exploit the feature information in the images, geometric dimension and semantic dimension enhancement modules are designed. The geometric module is designed from the perspective that the position information between neighboring object pairs will affect each other, which can improve the recall rate of the overall relationship in the dataset. The semantic module further processes the embedded word vector, which can enhance the acquisition of semantic information. Then, to improve the recall rate of the tail data, the Class Balanced Seesaw Loss (CBSLoss) is designed for the tail data. The recall rate of the prediction is improved by penalizing the body or tail relations that are judged incorrectly in the dataset. The experimental findings demonstrate that the GSI method performs better than mainstream models in terms of the mean Recall@K (mR@K) metric in three tasks. The long-tailed imbalance in the Visual Genome 150 (VG150) dataset is addressed better using the GSI method than by most of the existing methods.

• Journal of Biomedical Engineering Research
• /
• v.40 no.2
• /
• pp.62-67
• /
• 2019

The purpose of this study was to evaluate the performance of deep neural network model in order to determine whether there is a risk factor for coronary artery disease based on the cardiac variation parameter. The study used unidentifiable 297 data to evaluate the performance of the model. Input data consists of heart rate parameters, which are SDNN (standard deviation of the N-N intervals), PSI (physical stress index), TP (total power), VLF (very low frequency), LF (low frequency), HF (high frequency), RMSSD (root mean square of successive difference) APEN (approximate entropy) and SRD (successive R-R interval difference), the age group and sex. Output data are divided into normal and patient groups, and the patient group consists of those diagnosed with diabetes, high blood pressure, and hyperlipidemia among the various risk factors that can cause coronary artery disease. Based on this, a binary classification model was applied using Deep Neural Network of deep learning techniques to classify normal and patient groups efficiently. To evaluate the effectiveness of the model used in this study, Kernel SVM (support vector machine), one of the classification models in machine learning, was compared and evaluated using same data. The results showed that the accuracy of the proposed deep neural network was train set 91.79% and test set 85.56% and the specificity was 87.04% and the sensitivity was 83.33% from the point of diagnosis. These results suggest that deep learning is more efficient when classifying these medical data because the train set accuracy in the deep neural network was 7.73% higher than the comparative model Kernel SVM.

• Journal of Korean Society for Geospatial Information Science
• /
• v.14 no.4 s.38
• /
• pp.19-26
• /
• 2006

This study assessed the accuracies of the ground information extracting methods from the LiDAR data. Especially, it compared two kinds of method, one of them is using directly the raw LiDAR data which is point type vector data and the other is using changed data to DSM type as the normal grid type. The methods using Local Maxima and Entropy methods are applied as a former case, and for the other case, this study applies the method using edge detection with filtering and the generated reference surface by the mean filtering. Then, the accuracy assessment are performed with these results, DEM constructed manually and the error permitted limit in scale of digital map. As a results, each DEM mean errors of methods using edge detection with filtering, reference surface, Local Maxima and Entropy are 0.27m, 2.43m, 0.13m and 0.10m respectively. Hence, the method using entropy presented the highest accuracy. And an accuracy from a method directly using the raw LiDAR data has higher accuracy than the method using changed data to DSM type relatively.

• The Journal of the Acoustical Society of Korea
• /
• v.16 no.5
• /
• pp.112-121
• /
• 1997

In this paper, we study a model parameter compensation method for noise-robust speech recognition. We study model parameter compensation on a sentence by sentence and no other informations are used. Parallel model combination(PMC), well known as a model parameter compensation algorithm, is implemented and used for a reference of performance comparision. We also propose a modified PMC method which tunes model parameter with an association factor that controls average variability of gaussian mixtures and variability of single gaussian mixture per state for more robust modeling. We obtain a re-estimation solution of environmental variables based on the expectation-maximization(EM) algorithm in the cepstral domain. To evaluate the performance of the model compensation methods, we perform experiments on speaker-independent isolated word recognition. Noise sources used are white gaussian and driving car noise. To get corrupted speech we added noise to clean speech at various signal-to-noise ratio(SNR). We use noise mean and variance modeled by 3 frame noise data. Experimental result of the VTS approach is superior to other methods. The scheme of the zero order VTS approach is similar to the modified PMC method in adapting mean vector only. But, the recognition rate of the Zero order VTS approach is higher than PMC and modified PMC method based on log-normal approximation.

• The Journal of Korean Society for Radiation Therapy
• /
• v.23 no.1
• /
• pp.31-39
• /
• 2011

Purpose: Helical Tomotherapy allows only coplanar beam delivery because it does not allow couch rotation. We investigated a method to introduce non-coplanar beam by tilting a patient's head for Tomotherapy. The aim of this study was to compare intrafractional movement during Tomotherapy between coplanar and non-coplanar patient's setup. Materials and Methods: Helical Tomotherapy was used for treating eight patients with intracranial tumor. The subjects were divided into three groups: one group (coplanar) of 2 patients who lay on S-plate with supine position and wore thermoplastic mask for immobilizing the head, second group (non-coplanar) of 3 patients who lay on S-plate with supine position and whose head was tilted with Variable Axis Baseplate and wore thermoplastic mask, and third group (non-coplanar plus mouthpiece) of 3 patients whose head was tilted and wore a mouthpiece immobilization device and thermoplastic mask. The patients were treated with Tomotherapy after treatment planning with Tomotherapy Planning System. Megavoltage computed tomography (MVCT) was performed before and after treatment, and the intrafractional error was measured with lateral(X), longitudinal(Y), vertical(Z) direction movements and vector ($\sqrt{x^2+y^2+z^2}$) value for assessing overall movement. Results: Intrafractional error was compared among three groups by taking the error of MVCT taken after the treatment. As the correction values (X, Y, Z) between MVCT image taken after treatment and CT-simulation image are close to zero, the patient movement is small. When the mean values of movement of each direction for non-coplanar setup were compared with coplanar setup group, X-axis movement was decreased by 13%, but Y-axis and Z-axis movement were increased by 109% and 88%, respectively. Movements of Y-axis and Z-axis with non-coplanar setup were relatively greater than that of X-axis since a tilted head tended to slip down. The mean of X-axis movement of the group who used a mouthpiece was greater by 9.4% than the group who did not use, but the mean of Y-axis movement was lower by at least 64%, and the mean of Z-axis was lower by at least 67%, and the mean of Z-axis was lower by at least 67%, and the vector was lower by at least 59% with the use of a mouthpiece. Among these 8 patients, one patient whose tumor was located on left frontal lobe and left basal ganglia received reduced radiation dose of 38% in right eye, 23% in left eye, 30% in optic chiasm, 27% in brain stem, and 8% in normal brain with non-coplanar method. Conclusion: Tomotherapy only allows coplanar delivery of IMRT treatment. To complement this shortcoming, Tomotherapy can be used with non-coplanar method by artificially tilting the patient's head and using an oral immobilization instrument to minimize the movement of patient, when intracranial tumor locates near critical organs or has to be treated with high dose radiation.

• Journal of the Korea Society of Computer and Information
• /
• v.16 no.12
• /
• pp.187-196
• /
• 2011

Induction motors play a vital role in aeronautical and automotive industries so that many researchers have studied on developing a fault detection and classification system of an induction motor to minimize economical damage caused by its fault. With this reason, this paper extracts robust feature vectors from the normal/abnormal vibration signals of the induction motor in noise circumstance: partial autocorrelation (PARCOR) coefficient, log spectrum powers (LSP), cepstrum coefficients mean (CCM), and mel-frequency cepstrum coefficient (MFCC). Then, we classified different types of faults of the induction motor by using the extracted feature vectors as inputs of a neural network. To find optimal feature vectors, this paper evaluated classification performance with 2 to 20 different feature vectors. Experimental results showed that five to six features were good enough to give almost 100% classification accuracy except features by CCM. Furthermore, we considered that vibration signals could include noise components caused by surroundings. Thus, we added white Gaussian noise to original vibration signals, and then evaluated classification performance. The evaluation results yielded that LSP was the most robust in noise circumstance, then PARCOR and MFCC followed by LSP, respectively.

• Proceedings of the KSMRM Conference
• /
• 2001.11a
• /
• pp.111-111
• /
• 2001

목적： To evaluate the feasibility of the event-related functional MR study using power grip studying the hand motor system 대상 및 방법： Event-related functional MRI was performed on a 1.5T MR unit in seven norm volunteers (man=7, right-handedness=2, left-handedness=5, mean age： 25 years). A single-shot GRE-EPI sequence (TR/TE/flip angle： 1000ms/40ms/90, FOV = 240 mm matrix= 64$\times$64, slice thickness/gap = 5mm/0mm, 7 true axial slices) was used for functiona MR images. A flow-sensitive conventional gradient echo sequence (TR/TE/flip angl 50ms/4ms/60) was used for high-resolution anatomical images. To minimize the gross hea motion, neck-holders (MJ-200, USA) were used. A series of MR images were obtained in axial planes covering motor areas. To exclude motion-corrupted images, all MR images wer surveyed in a movie procedure and evaluated using the estimation of center of mass of ima signal intensities. Power grip task consisted of the powerful grip of all right fingers and hand movement ta used very fast right finger tapping at a speed of 3 per 1 second. All tasks were visual-guid by LCD projector (SHARP, Japan). Two tasks consisted of 134 phases including 7 activatio and 8 rest periods. Active stimulations were performed during 2 seconds and rest period were 15 seconds and total scan time per one task was 2 min 14 sec. Statistical maps we obtained using cross-correlation method. Reference vector was time-shifted by 4 seconds an Gaussian convolution with a FWHM of 4 seconds was applied to it. The threshold in p val for the activation sites was set to be 0.001. All mapping procedures were peformed usin homemade program an IDL (Research Systems Inc., USA) platform. We evaluated the activation patterns of the motor system of power grip compared to hand movement in t event-related functional MRI.

• Archives of Plastic Surgery
• /
• v.36 no.1
• /
• pp.1-10
• /
• 2009

Purpose: Partial necrosis of skin flaps remains a substantial problem in reconstructive surgery. We investigated the potential use of an adenovirus vector encoding the VEGF, COMP-angiopoietin-1 gene in an attempt to promote the viability of the inferior epigastric artery flap in a rat model. Methods: Three by six cm lower abdominal transverse skin flaps, supplied only by the left inferior epigastric artery, were designed. After skin flap elevation, the adenovirus VEGF and adenovirus COMP-angiopoietin-1 were injected into the distal portion of the flap, which has a high tendency of developing flap ischemia. Control animals were injected with the same volume of normal saline. On 3, 7 and 14 days after the flap elevation, the flap survival and vascularization were assessed using Visitrak digital$^{(R)}$, CD31 immunohistochemistry in addition to evaluating the general histological characteristics. Results: There was a significant increase in the mean percentage of flap viability by 89.8%, 91.1% and 94.8% in flaps transfected with adenovirus VEGF, COMP-angiopoietin-1, coadministraion of VEGF and COMP-angiopoietin-1 at seven days, and by 95.6%, 94.8% and 96.3% at 14 days. Histological assessment revealed that there were more blood vessels formed after adenovirus with VEGF, COMP-angiopoietin-1 or VEGF plus COMP-angiopoietin-1 than with adenovirus Lac Z. Conclusion: The results of this study suggest that adenovirus-mediated VEGF, COMP-angiopoietin-1 gene therapy, promote therapeutic angiogenesis in patients that undergo reconstructive procedures.

• KIPS Transactions on Software and Data Engineering
• /
• v.10 no.3
• /
• pp.99-108
• /
• 2021

There are unrestricted conditions on the driver's face inside the vehicle, such as changes in lighting, partial occlusion and various changes in the driver's condition. In this paper, we propose a face identification system in an unrestricted vehicle environment. The proposed method uses a near-infrared (NIR) camera to minimize the changes in facial images that occur according to the illumination changes inside and outside the vehicle. In order to process a face exposed to extreme light, the normal face image is changed to a simulated overexposed image using mean and variance for training. Thus, facial classifiers are simultaneously generated under both normal and extreme illumination conditions. Our method identifies a face by detecting facial landmarks and aggregating the confidence score of each landmark for the final decision. In particular, the performance improvement is the highest in the class where the driver wears glasses or sunglasses, owing to the robustness to partial occlusions by recognizing each landmark. We can recognize the driver by using the scores of remaining visible landmarks. We also propose a novel robust rejection and a new evaluation method, which considers the relations between registered and unregistered drivers. The experimental results on our dataset, PolyU and ORL datasets demonstrate the effectiveness of the proposed method.

• The Journal of Korean Society for Radiation Therapy
• /
• v.30 no.1_2
• /
• pp.97-105
• /
• 2018

Purpose : Proton Therapy using Bragg-peak, because it has distinct characteristics in providing maximum dosage for tumor and minimal dosage for normal tissue, a medical imaging system that can quantify changes in patient position or treatment area is of paramount importance to the treatment of protons. The purpose of this research is to evaluate the usefulness of the algorithm by comparing the image matching through the set-up and in-house code through the existing dips program by producing a Matlab-based in-house registration code to determine the error value between dips and DRR to evaluate the accuracy of the existing treatment. Materials and Methods : Thirteen patients with brain tumors and head and neck cancer who received proton therapy were included in this study and used the DIPS Program System (Version 2.4.3, IBA, Belgium) for image comparison and the Eclipse Proton Planning System (Version 13.7, Varian, USA) for patient treatment planning. For Validation of the Registration method, a test image was artificially rotated and moved to match the existing image, and the initial set up image of DIPS program of existing set up process was image-matched with plan DRR, and the error value was obtained, and the usefulness of the algorithm was evaluated. Results : When the test image was moved 0.5, 1, and 10 cm in the left and right directions, the average error was 0.018 cm. When the test image was rotated counterclockwise by 1 and $10^{\circ}$, the error was $0.0011^{\circ}$. When the initial images of four patients were imaged, the mean error was 0.056, 0.044, and 0.053 cm in the order of x, y, and z, and 0.190 and $0.206^{\circ}$ in the order of rotation and pitch. When the final images of 13 patients were imaged, the mean differences were 0.062, 0.085, and 0.074 cm in the order of x, y, and z, and 0.120 cm as the vector value. Rotation and pitch were 0.171 and $0.174^{\circ}$, respectively. Conclusion : The Matlab-based In-house Registration code produced through this study showed accurate Image matching based on Intensity as well as the simple image as well as anatomical structure. Also, the Set-up error through the DIPS program of the existing treatment method showed a very slight difference, confirming the accuracy of the proton therapy. Future development of additional programs and future Intensity-based Matlab In-house code research will be necessary for future clinical applications.