• Journal of Internet Computing and Services
• /
• v.24 no.6
• /
• pp.91-98
• /
• 2023

Recently, according to development of artificial intelligence, a wide range of industry being automatic and optimized. Also we can find out some research of using supervised learning for deteceting defect of railway in domestic rail industry. However, there are structures other than rails on the track, and the fastener is a device that binds the rail to other structures, and periodic inspections are required to prevent safety accidents. In this paper, we present a method of reducing cost for labeling using semi-supervised and transfer model trained on rail fastener data. We use Resnet50 as the backbone network pretrained on ImageNet. At first we randomly take training data from unlabeled data and then labeled that data to train model. After predict unlabeled data by trained model, we adopted a method of adding the data with the highest probability for each class to the training data by a predetermined size. Futhermore, we also conducted some experiments to investigate the influence of the number of initially labeled data. As a result of the experiment, model reaches 92% accuracy which has a performance difference of around 5% compared to supervised learning. This is expected to improve the performance of the classifier by using relatively few labels without additional labeling processes through the proposed method.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.20 no.8
• /
• pp.579-585
• /
• 2019

Increasing the detection probability of underwater targets necessitates securing the towing stability of the active towed SONAR. In this paper, to confirm the effects of tail wing fin on towing attitude and towing stability, two scale model experiments and one sea trials were conducted and the results were analyzed. The scale model tests measured the towing behavior of each of the tail fin shapes according to towing speed in a towing tank. The shape of the tail fin used in the scale model test was tested with an I-type tail fine and four Y-type tail fins, totaling five tail fins of the two kinds. The first scale model test confirmed that the Y-type tail fin was superior to the I-type tail fin in towing attitude and towing stability. The second scale model test confirmed the characteristics of the vertical tail fin height increase and the lower horizontal tail fin inclination angle application shape based on the Y-type tail fin. The shape of the application of the lower horizontal tail fin inclination angle showed the best performance. In order to verify the results of the scale model test, a full size model was constructed, sea trials were performed, and the towing attitude was measured. The results were similar to those of the scale model test.

• Journal of the Korean Society of Radiology
• /
• v.81 no.3
• /
• pp.610-619
• /
• 2020

Purpose The study aimed to investigate the role of Prostate Imaging Reporting and Data System version 2 (PI-RADS v2) in predicting incidental prostate cancer (PCa) or urothelial carcinoma (UCa) extension in urinary bladder (UB) cancer patients. Materials and Methods A total of 72 UB cancer patients who underwent radical cystoprostatectomy and 3 Tesla multiparametric MRI before surgery were enrolled. PI-RADS v2 ratings were assigned by two independent radiologists. All prostate specimens were examined by a single pathologist. We compared the multiparametric MRI findings rated using PI-RADS v2 with the pathologic data. Results Of the 72 UB cancer patients, 29 had incidental PCa (40.3%) and 20 showed UCa extension (27.8%), with an overlap for 3 patients. With a score of 4 as the cut-off value for predicting incidental PCa, the diagnostic accuracy was 65.3%, specificity was 90.7%, and positive predictive value (PPV) was 66.7%. The diagnostic accuracy for incidental UCa extension was 47.2%, specificity was 92.3%, and PPV was 83.3%. Conclusion Despite the low diagnostic accuracy, the PPV and specificity were relatively high. Therefore, PI-RADS v2 scores of 1, 2, or 3 may help exclude the probability of incidental PCa or UCa extension.

• Korean Journal of Culture and Social Issue
• /
• v.14 no.4
• /
• pp.1-18
• /
• 2008

The present study examined the accuracy of polygraph tests through two types of statistical methods with unknown ground truth. One method evaluated the accuracy based on the rates of agreements between polygraph test results of crime suspects and prosecutors' indictment decisions for them. Those crime suspects were tested with polygraph by the Prosecutors' Office of the Republic of Korea between 2000 and 2004. The other method estimated the accuracy by using the latent class analysis based on the frequency distributions of the polygraph results and indictments during 2006. Excluding cases that were 'inconclusive' on the polygraph test, the study showed that the accuracy of the polygraph tests is .914 (SE=.004) for the 2000-2004 data, and .885 (SE=.021) for the 2006 data. With the inclusion of 'inconclusive' cases in the 2006 data, the results from the latent class analysis showed the accuracy in the range between .707 and .734 (SE=.027~.031), with false positives between .078 and .087 (SE=.019~.023), and false negatives between .029 and .078 (SE=.010~.023). The probability that the polygraph test correctly classifies subjects appeared to be in the range between .912 and .925 (SE=.013-.016) for those who lie, and in the range between .867 to .955 (SE=.011-.040) for those who tell the truth.

• Journal of the Institute of Electronics and Information Engineers
• /
• v.51 no.6
• /
• pp.201-208
• /
• 2014

In this paper, we propose the classifier structure design algorithm using DTW. Proposed algorithm uses DTW result to design the binary tree architecture based on the SVM which classify the multi-class data. Design the binary tree architecture for Support Vector Machine(SVM-BTA) using the threshold criterion calculated by the sum columns in square matrix which components are the reference data from each class. For comparison the performance of the proposed algorithm, compare the results of classifiers which binary tree structure are designed based on database and k-means algorithm. The data used for classification is 333 signals from 18 classes of underwater transient noise. The proposed classifier has been improved classification performance compared with classifier designed by database system, and probability of detection for non-biological transient signal has improved compare with classifiers using k-means algorithm. The proposed SVM-BTA classified 68.77% of biological sound(BO), 92.86% chain(CHAN) the mechanical sound, and 100% of the 6 kinds of the other classes.

• KSCE Journal of Civil and Environmental Engineering Research
• /
• v.26 no.1D
• /
• pp.195-202
• /
• 2006

The infiltration route analysis is a military application using geospatial information technology. The result of the analysis would present vulnerable routes for potential enemy infiltration. In order to find the susceptible routes, optimal path search algorithms (Dijkstra's and $A^*$) were used to minimize the cost function, summation of detection probability. The cost function was produced by capability of TOD (Thermal Observation Device), results of viewshed analysis using DEM (Digital Elevation Model) and two related geospatial information coverages (obstacle and vegetation) extracted from VITD (Vector product Interim Terrain Data). With respect to 50m by 50m cells, the individual cost was computed and recorded, and then the optimal infiltration routes was found while minimizing summation of the costs on the routes. The proposed algorithm was experimented in Daejeon region in South Korea. The test results show that Dijkstra's and $A^*$ algorithms do not present significant differences, but A* algorithm shows a better efficiency. This application can be used for both infiltration and surveillance. Using simulation of moving TOD, the most vulnerable routes can be detected for infiltration purpose. On the other hands, it can be inversely used for selection of the best locations of TOD. This is an example of powerful geospatial solution for military application.

• Food Science and Preservation
• /
• v.30 no.3
• /
• pp.459-471
• /
• 2023

The Bayesian network (BN) model was applied to analyze the characteristic variables that affect compliance with safety inspections of farmed eel during the production stage, using the data from 30,063 cases of eel aquafarm safety inspection in the Integrated Food Safety Information Network (IFSIN) from 2012 to 2021. The dataset for establishing the BN model included 77 non-conforming cases. Relevant HACCP data, geographic information about the aquafarms, and environmental data were collected and mapped to the IFSIN data to derive explanatory variables for nonconformity. Aquafarm HACCP certification, detection history of harmful substances during the last 5 y, history of nonconformity during the last 5 y, and the suitability of the aquatic environment as determined by the levels of total coliform bacteria and total organic carbon were selected as the explanatory variables. The highest achievable eel aquafarm noncompliance rate by manipulating the derived explanatory variables was 24.5%, which was 94 times higher than the overall farmed eel noncompliance rate reported in IFSIN between 2017 and 2021. The established BN model was validated using the IFSIN eel aquafarm inspection results conducted between January and August 2022. The noncompliance rate in the validation set was 0.22% (15 nonconformances out of 6,785 cases). The precision of BN model prediction was 0.1579, which was 71.4 times higher than the non-compliance rate of the validation set.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.23 no.7
• /
• pp.1797-1806
• /
• 1998

Future satellite communication systems will be developed at Ka-band (20/30 GHz) owing to the relatively wide frequency allocation and current freedom from terrestrial interference for multimedia services. A serious disadvantage of the Ka-band, however, is the very high atmospheric attenuation in rainy weather. Synchronous CDMA drastically redces the effect of self-noise with several interesting features of CDMA for mobile communications such as fixible freuqncy rese, the capability of performin soft-handover and a lower sensitivity to interference. This paper evaluates the performance of a synchronous CDMA reture link for a Ka-band geostationary satellite communication system. For a fixed satellite channel whose characteristics depend on weather conditions, the signal envelope and phase for this channel is modelled as Gaussian. The bit error and outage probability, and the detection loss due to imperfect chip timing synchronization is analytically evaluated and the system capacity degaradation due to the weather condition is estimated. The two cases consist of the general case in which all users are affected by rain condition, and the worst case in which the reference user is only affected by rain attenuation. the results for two cases of rain condition clearly show that synchronous CDMA eases the power control requirements and has less sensitivity to imperfect power control.

• Journal of the Institute of Convergence Signal Processing
• /
• v.23 no.3
• /
• pp.166-172
• /
• 2022

In this study, we photograph driving obstacle objects such as crosswalks, side spheres, manholes, braille blocks, partial ramps, temporary safety barriers, stairs, and inclined curb that hinder or cause inconvenience to the movement of the vulnerable using electric mobility aids. We develop an optimal AI model that classifies photographed objects and automatically recognizes them, and implement an algorithm that can efficiently determine obstacles in front of electric mobility aids. In order to enable object detection to be AI learning with high probability, the labeling form is labeled as a polygon form when building a dataset. It was developed using a Mask R-CNN model in Detectron2 framework that can detect objects labeled in the form of polygons. Image acquisition was conducted by dividing it into two groups: the general public and the transportation weak, and image information obtained in two areas of the test bed was secured. As for the parameter setting of the Mask R-CNN learning result, it was confirmed that the model learned with IMAGES_PER_BATCH: 2, BASE_LEARNING_RATE 0.001, MAX_ITERATION: 10,000 showed the highest performance at 68.532, so that the user can quickly and accurately recognize driving risks and obstacles.

• Journal of the Institute of Convergence Signal Processing
• /
• v.24 no.2
• /
• pp.119-125
• /
• 2023

In this study, we photograph driving obstacle objects such as crosswalks, side spheres, manholes, braille blocks, partial ramps, temporary safety barriers, stairs, and inclined curb that hinder or cause inconvenience to the movement of the vulnerable using electric mobility aids. We develop an optimal AI model that classifies photographed objects and automatically recognizes them, and implement an algorithm that can efficiently determine obstacles in front of electric mobility aids. In order to enable object detection to be AI learning with high probability, the labeling form is labeled as a polygon form when building a dataset. It was developed using a Mask R-CNN model in Detectron2 framework that can detect objects labeled in the form of polygons. Image acquisition was conducted by dividing it into two groups: the general public and the transportation weak, and image information obtained in two areas of the test bed was secured. As for the parameter setting of the Mask R-CNN learning result, it was confirmed that the model learned with IMAGES_PER_BATCH: 2, BASE_LEARNING_RATE 0.001, MAX_ITERATION: 10,000 showed the highest performance at 68.532, so that the user can quickly and accurately recognize driving risks and obstacles.