• Journal of Intelligence and Information Systems
• /
• v.28 no.4
• /
• pp.329-346
• /
• 2022

Recently, as the market for short form videos (Instagram, TikTok, YouTube) on social media has gradually increased, research using them is actively being conducted in the artificial intelligence field. A representative research field is Video to Shop, which detects fashion products in videos and searches for product images. In such a video-based artificial intelligence model, product features are extracted using convolution operations. However, due to the limitation of computational resources, extracting features using all the frames in the video is practically impossible. For this reason, existing studies have improved the model's performance by sampling only a part of the entire frame or developing a sampling method using the subject's characteristics. In the existing Video to Shop study, when sampling frames, some frames are randomly sampled or sampled at even intervals. However, this sampling method degrades the performance of the fashion product search model while sampling noise frames where the product does not exist. Therefore, this paper proposes a sampling method MF (Missing Fashion items on frame) sampler that removes noise frames and improves the performance of the search model. MF sampler has improved the problem of resource limitations by developing a keyframe mechanism. In addition, the performance of the search model is improved through noise frame removal using the noise detection model. As a result of the experiment, it was confirmed that the proposed method improves the model's performance and helps the model training to be effective.

• Wind and Structures
• /
• v.36 no.4
• /
• pp.237-247
• /
• 2023

The emergence of high-rise buildings has necessitated frequent structural health monitoring and maintenance for safety reasons. Wind causes damage and structural changes on tall structures; thus, safe structures should be designed. The pressure developed on tall buildings has been utilized in previous research studies to assess the impacts of wind on structures. The wind tunnel test is a primary research method commonly used to quantify the aerodynamic characteristics of high-rise buildings. Wind pressure is measured by placing pressure sensor taps at different locations on tall buildings, and the collected data are used for analysis. However, sensors may malfunction and produce erroneous data; these data losses make it difficult to analyze aerodynamic properties. Therefore, it is essential to generate missing data relative to the original data obtained from neighboring pressure sensor taps at various intervals. This study proposes a deep learning-based, deep convolutional generative adversarial network (DCGAN) to restore missing data associated with faulty pressure sensors installed on high-rise buildings. The performance of the proposed DCGAN is validated by using a standard imputation model known as the generative adversarial imputation network (GAIN). The average mean-square error (AMSE) and average R-squared (ARSE) are used as performance metrics. The calculated ARSE values by DCGAN on the building model's front, backside, left, and right sides are 0.970, 0.972, 0.984 and 0.978, respectively. The AMSE produced by DCGAN on four sides of the building model is 0.008, 0.010, 0.015 and 0.014. The average standard deviation of the actual measures of the pressure sensors on four sides of the model were 0.1738, 0.1758, 0.2234 and 0.2278. The average standard deviation of the pressure values generated by the proposed DCGAN imputation model was closer to that of the measured actual with values of 0.1736,0.1746,0.2191, and 0.2239 on four sides, respectively. In comparison, the standard deviation of the values predicted by GAIN are 0.1726,0.1735,0.2161, and 0.2209, which is far from actual values. The results demonstrate that DCGAN model fits better for data imputation than the GAIN model with improved accuracy and fewer error rates. Additionally, the DCGAN is utilized to estimate the wind pressure in regions of buildings where no pressure sensor taps are available; the model yielded greater prediction accuracy than GAIN.

• International Journal of Highway Engineering
• /
• v.16 no.3
• /
• pp.119-129
• /
• 2014

PURPOSES : This investigational survey is to observe a proper spatial aggregation method for path travel time estimation using the hi-pass DSRC system. METHODS : The links which connect the nodes of section detectors location are used for path travel time estimation traditionally. It makes some problem such as increasing accumulation errors and processing times. In this background, the new links composition methods for spatial aggregation are considered by using some types of nodes as IC, JC, RSE combination. Path travel times estimated by new aggregation methods are compared with PBM travel times by MAE, MAPE and statistical hypothesis tests. RESULTS : The results of minimum sample size and missing rate for 5 minutes aggregation interval are satisfied except for JC link path travel time in Seoul TG~Kuemho JC. Thus, it was additionally observed for minimum sample size satisfaction. In 15, 30 minutes and 1 hour aggregation intervals, all conditions are satisfied by the minimum sample size criteria. For accuracy test and statistical hypothesis test, it has been proved that RSE, Conzone, IC, JC links have equivalent errors and statistical characteristics. CONCLUSIONS : There are some errors between the PBM and the LBM methods that come from dropping vehicles by rest areas. Consequently, this survey result means each of links compositions are available for the estimation of path travel time when PBM vehicles are missed.

• The Korean Journal of Applied Statistics
• /
• v.22 no.6
• /
• pp.1177-1190
• /
• 2009

The traffic operating speed is one of important information to measure a road capacity. When we supply the information of the road of high traffic by using navigation, offering the present traffic information and the forecasted future information are the outstanding functions to serve the more accurate expected times and intervals. In this study, we proposed the traffic speed forecasting model using the accumulated traffic speed data of the road and highway and forecasted the average speed for each the road and high interval and each time interval using Fourier transformation and time series regression model with trigonometrical function. We also propose the proper method of missing data imputation and treatment for the outliers to raise an accuracy of the traffic speed forecasting and the speed grouping method for which data have similar traffic speed pattern to increase an efficiency of analysis.

• Journal of Korea Spatial Information System Society
• /
• v.7 no.2 s.14
• /
• pp.67-79
• /
• 2005

For tracing tag locations, the trajectories should be modeled and indexed in a radio frequency identification (RFID) system. The trajectory of a tag is represented as a line that connects two spatiotemporal locations captured when the tag enters and leaves the vicinity of a reader. If a tag enters but does not leave a reader, its trajectory is represented only as a point captured at entry. Because the information that a tag stays in a reader is missing from the trajectory represented only as a point, it is impossible to find the tag that remains in a reader. To solve this problem we propose the data model in which trajectories are defined as intervals and new index scheme called the Interval R-tree. We also propose new insert and split algorithms to enable efficient query processing. We evaluate the performance of the proposed index scheme and compare it with the R-tree and the R*-tree. Our experiments show that the new index scheme outperforms the other two in processing queries of tags on various datasets.

• Journal of KIISE:Databases
• /
• v.33 no.1
• /
• pp.56-68
• /
• 2006

For tracing tag locations, the trajectories should be modeled and indexed in radio frequency identification (RFID) systems. The trajectory of a tag can be represented as a line that connects two spatiotemporal locations captured when the tag enters and leaves the vicinity of a reader. If a tag enters but does not leave a reader, its trajectory is represented only as a point captured at entry. Because the information that a tag stays in a reader is missing from the trajectory represented only as a point, it is impossible to find the tag that remains in a reader. To solve this problem we propose the data model in which trajectories are defined as time-parameterized intervals and new index scheme called the Time Parameterized Interval R-tree. We also propose new insert and split algorithms that reduce the area of nodes to enable efficient query processing. We evaluate the performance of the proposed index scheme and compare it with previous indexes on various datasets.

• Journal of Biomedical Engineering Research
• /
• v.42 no.4
• /
• pp.175-185
• /
• 2021

Purpose: Vital sign are used to help assess the general physical health of a person, give clues to possible diseases, and show progress toward recovery. Researchers are using vital sign data and AI(artificial intelligence) to manage a variety of diseases and predict mortality. In order to analyze vital sign data using AI, it is important to select and extract vital sign data suitable for research purposes. Methods: We developed a method to visualize vital sign and early warning scores by processing retrospective vital sign data collected from EMR(electronic medical records) and patient monitoring devices. The vital sign data used for development were obtained using the open EMR big data MIMIC-III and the wearable patient monitoring device(CareTaker). Data processing and visualization were developed using Python. We used the development results with machine learning to process the prediction of mortality in ICU patients. Results: We calculated NEWS(National Early Warning Score) to understand the patient's condition. Vital sign data with different measurement times and frequencies were sampled at equal time intervals, and missing data were interpolated to reconstruct data. The normal and abnormal states of vital sign were visualized as color-coded graphs. Mortality prediction result with processed data and machine learning was AUC of 0.892. Conclusion: This visualization method will help researchers to easily understand a patient's vital sign status over time and extract the necessary data.

• The Journal of the Convergence on Culture Technology
• /
• v.8 no.6
• /
• pp.899-904
• /
• 2022

As the world becomes an aging society due to a decrease in the birth rate and an increase in life expectancy, a system for health management of the elderly population is needed. Among them, various studies on occupancy and activity types are being conducted for smart home care services for indoor health management. In this paper, we propose a random forest model that classifies activity type as well as occupancy status through indoor temperature and humidity, CO2, fine dust values and UWB radar positioning for smart home care service. The experiment measures indoor environment and occupant positioning data at 2-second intervals using three sensors that measure indoor temperature and humidity, CO2, and fine dust and two UWB radars. The measured data is divided into 80% training set data and 20% test set data after correcting outliers and missing values, and the random forest model is applied to evaluate the list of important variables, accuracy, sensitivity, and specificity.

• Journal of Life Science
• /
• v.28 no.11
• /
• pp.1255-1261
• /
• 2018

Whole genome analysis have been made possible with the development of DNA sequencing technologies and discovery of many single nucleotide polymorphisms (SNPs). Large number of SNP can be analyzed with SNP chips, since SNPs of human as well as livestock genomes are available. Among the various missing nucleotide imputation programs, Minimac3 software is suggested to be highly accurate, with a simplified workflow and relatively fast. In the present study, we used Minimac3 program to perform genomic missing value substitution 1,226 animals 770K SNP chip and imputing missing SNPs with next generation sequencing data from 311 animals. The accuracy on each chromosome was about 94~96%, and individual sample accuracy was about 92~98%. After imputation of the genotypes, SNPs with R Square ($R^2$) values for three conditions were 0.4, 0.6, and 0.8 and the percentage of SNPs were 91%, 84%, and 70% respectively. The differences in the Minor Allele Frequency gave $R^2$ values corresponding to seven intervals (0, 0.025), (0.025, 0.05), (0.05, 0.1), (0.1, 0.2), (0.2, 0.3). (0.3, 0.4) and (0.4, 0.5) of 64~88%. The total analysis time was about 12 hr. In future SNP chip studies, as the size and complexity of the genomic datasets increase, we expect that genomic imputation using Minimac3 can improve the reliability of chip data for Hanwoo discrimination.

• Magazine of the Korean Society of Agricultural Engineers
• /
• v.21 no.3
• /
• pp.83-91
• /
• 1979

This study was conducted in order to find out the performance of rice transplanters in accordance with the change of the trans-planting days after pudding and the water depth flooding the paddy field at the time of transplanting : and thus to select the optimum paddy field preparation procedures for an efficient utilization of rice transplanters. The performance factors of the two different types of rice transplanters were measured during the first 6 consecutive days after puddling and with 3 different levels of water depth flooding the paddy fields. The results of this study were analysed and summarized as follows : 1. Wheel sinkage decreased very rapidly from 0 to 2 days after puddling and slowly from 3 to 5 days after puddling. 2. The depth of the test cone penetration decreased rapidly during the first few days after puddling. It was 17.8cm just after puddling, and decreased to 13.4cm one day after puddling. After 2 days, the rate of decrease was dampened, and after 5 days it kept constant value of 9.2cm. 3. Two days after puddling, the hill interval was 15.8cm (98.75% of the preset value) for broadcasted seedling rice transplanter with 3cm flooding depth : This value was the closest to the pre-adjusted value of 16cm. The general performance of broadcasted-seedling type rice transplanter was better than that of strip-seedling type rice transplanter. 4. Usually the working performance of a rice transplanter is evaluated with uniformity and adjustability of the hill intervals. The hill interval was the most uniform and closest to the pre-set value of 16cm when planted two days after puddling with 3cm of water depth. When it was inavoidable to plant 4 days after puddling with stripseedling type rice transplanter, it is advisable to let the water flooded somewhat deeper. 5. The percentage of missing hills including floating and burried seedlings was the highest just after puddling and ie decreased substancially until 3 days after puddling and then it increased again. Hence, the optimal time transplanting is to be between 2 and 3 days after puddling. 6. Better postures of planted seedlings were found when planter 2 days after puddling than 3 days after puddling. Six cm of flooding water depth always gave the best results with respect to the postures of planted seedlings. Broadcasted-seedling rice transplanter, in general, showed better posture of planted seedlings than did strip-seedling type rice transplanter. 7. Judging from the above results, the optimal conditions will be 3cm of flooding depth and transplanting between 2 and 3 days after puddling.