• Asia-pacific Journal of Multimedia Services Convergent with Art, Humanities, and Sociology
• /
• v.7 no.3
• /
• pp.329-337
• /
• 2017

This paper proposes a search method that explores parameters C and σ values of support vector machines (SVM) to improve performance while maintaining search accuracy. A traditional grid search method requires tremendous computational times because it searches all available combinations of C and σ values to find optimal combinations which provide the best performance of SVM. To address this issue, this paper proposes a deep search method that reduces computational time. In the first stage, it divides C-σ- accurate metrics into four regions, searches a median value of each region, and then selects a point of the highest accurate value as a start point. In the second stage, the selected start points are re-divided into four regions, and then the highest accurate point is assigned as a new search point. In the third stage, after eight points near the search point. are explored and the highest accurate value is assigned as a new search point, corresponding points are divided into four parts and it calculates an accurate value. In the last stage, it is continued until an accurate metric value is the highest compared to the neighborhood point values. If it is not satisfied, it is repeated from the second stage with the input level value. Experimental results using normal and defect bearings show that the proposed deep search algorithm outperforms the conventional algorithms in terms of performance and search time.

• Smart Media Journal
• /
• v.12 no.1
• /
• pp.9-16
• /
• 2023

Recently, a convolutional neural network (CNN) based system is being developed to overcome the limitations of human resources in the apple quality classification of farmhouse. However, since convolutional neural networks receive only images of the same size, preprocessing such as sampling may be required, and in the case of oversampling, information loss of the original image such as image quality degradation and blurring occurs. In this paper, in order to minimize the above problem, to generate a image patch based graph of an original image and propose a random walk-based positional encoding method to apply the graph transformer model. The above method continuously learns the position embedding information of patches which don't have a positional information based on the random walk algorithm, and finds the optimal graph structure by aggregating useful node information through the self-attention technique of graph transformer model. Therefore, it is robust and shows good performance even in a new graph structure of random node order and an arbitrary graph structure according to the location of an object in an image. As a result, when experimented with 5 apple quality datasets, the learning accuracy was higher than other GNN models by a minimum of 1.3% to a maximum of 4.7%, and the number of parameters was 3.59M, which was about 15% less than the 23.52M of the ResNet18 model. Therefore, it shows fast reasoning speed according to the reduction of the amount of computation and proves the effect.

• 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.

• Journal of the Institute of Convergence Signal Processing
• /
• v.24 no.1
• /
• pp.27-33
• /
• 2023

Recently, water shortages have occurred due to climate change, and the need for water management of agricultural water has increased due to the occurrence of algal blooms in reservoirs. Existing algae prevention is operated by putting many people on site and misses the optimal spraying time due to movement through boats. In order to solve this problem, it is necessary to block contamination in advance and move within time to uniformly spray complex microorganisms uniformly. Control drones are used for pesticide spraying and can be applied to algae prevention work by utilizing control drones. In this paper, basic research for the establishment of a marine control system was conducted for application to the reservoir environment, and as one of the results, the characteristics of a drone nozzle, a core technology that can be used for control drones, were calculated. In particular, it was found that the existing agricultural control drones had a disadvantage that the concentration was non-uniform within the suggested spraying interval, and to compensate for this, nozzle positioning and nozzle spraying uniformity were calculated. Based on the experimental results, we develop a core algorithm for establishing an algal bloom monitoring system in the reservoir environment and propose a precision control technology that can be used for marine control work in the future.

• Journal of Intelligence and Information Systems
• /
• v.29 no.4
• /
• pp.325-345
• /
• 2023

In the rapidly evolving domain of e-commerce, our study presents a cohesive approach to enhance customer satisfaction prediction from online reviews, aligning methodological innovation with practical insights. We integrate the RFE-SHAP feature selection with LDA topic modeling to streamline predictive analytics in e-commerce. This integration facilitates the identification of key features-specifically, narrowing down from an initial set of 28 to an optimal subset of 14 features for the Random Forest algorithm. Our approach strategically mitigates the common issue of overfitting in models with an excess of features, leading to an improved accuracy rate of 84% in our Random Forest model. Central to our analysis is the understanding that certain aspects in review content, such as quality, fit, and durability, play a pivotal role in influencing customer satisfaction, especially in the clothing sector. We delve into explaining how each of these selected features impacts customer satisfaction, providing a comprehensive view of the elements most appreciated by customers. Our research makes significant contributions in two key areas. First, it enhances predictive modeling within the realm of e-commerce analytics by introducing a streamlined, feature-centric approach. This refinement in methodology not only bolsters the accuracy of customer satisfaction predictions but also sets a new standard for handling feature selection in predictive models. Second, the study provides actionable insights for e-commerce platforms, especially those in the clothing sector. By highlighting which aspects of customer reviews-like quality, fit, and durability-most influence satisfaction, we offer a strategic direction for businesses to tailor their products and services.

• Journal of the Korean Institute of Landscape Architecture
• /
• v.52 no.2
• /
• pp.96-109
• /
• 2024

The purpose of this study is to propose a method for evaluating the similarity of Show gardens using Deep Learning models, specifically VGG-16 and ResNet50. A model for judging the similarity of show gardens based on VGG-16 and ResNet50 models was developed, and was referred to as DRG (Deep Recognition of similarity in show Garden design). An algorithm utilizing GAP and Pearson correlation coefficient was employed to construct the model, and the accuracy of similarity was analyzed by comparing the total number of similar images derived at 1st (Top1), 3rd (Top3), and 5th (Top5) ranks with the original images. The image data used for the DRG model consisted of a total of 278 works from the Le Festival International des Jardins de Chaumont-sur-Loire, 27 works from the Seoul International Garden Show, and 17 works from the Korea Garden Show. Image analysis was conducted using the DRG model for both the same group and different groups, resulting in the establishment of guidelines for assessing show garden similarity. First, overall image similarity analysis was best suited for applying data augmentation techniques based on the ResNet50 model. Second, for image analysis focusing on internal structure and outer form, it was effective to apply a certain size filter (16cm × 16cm) to generate images emphasizing form and then compare similarity using the VGG-16 model. It was suggested that an image size of 448 × 448 pixels and the original image in full color are the optimal settings. Based on these research findings, a quantitative method for assessing show gardens is proposed and it is expected to contribute to the continuous development of garden culture through interdisciplinary research moving forward.

• Korean Journal of Radiology
• /
• v.22 no.6
• /
• pp.983-993
• /
• 2021

Objective: To investigate the image quality of ultralow-dose CT (ULDCT) of the chest reconstructed using a cycle-consistent generative adversarial network (CycleGAN)-based deep learning method in the evaluation of pulmonary tuberculosis. Materials and Methods: Between June 2019 and November 2019, 103 patients (mean age, 40.8 ± 13.6 years; 61 men and 42 women) with pulmonary tuberculosis were prospectively enrolled to undergo standard-dose CT (120 kVp with automated exposure control), followed immediately by ULDCT (80 kVp and 10 mAs). The images of the two successive scans were used to train the CycleGAN framework for image-to-image translation. The denoising efficacy of the CycleGAN algorithm was compared with that of hybrid and model-based iterative reconstruction. Repeated-measures analysis of variance and Wilcoxon signed-rank test were performed to compare the objective measurements and the subjective image quality scores, respectively. Results: With the optimized CycleGAN denoising model, using the ULDCT images as input, the peak signal-to-noise ratio and structural similarity index improved by 2.0 dB and 0.21, respectively. The CycleGAN-generated denoised ULDCT images typically provided satisfactory image quality for optimal visibility of anatomic structures and pathological findings, with a lower level of image noise (mean ± standard deviation [SD], 19.5 ± 3.0 Hounsfield unit [HU]) than that of the hybrid (66.3 ± 10.5 HU, p < 0.001) and a similar noise level to model-based iterative reconstruction (19.6 ± 2.6 HU, p > 0.908). The CycleGAN-generated images showed the highest contrast-to-noise ratios for the pulmonary lesions, followed by the model-based and hybrid iterative reconstruction. The mean effective radiation dose of ULDCT was 0.12 mSv with a mean 93.9% reduction compared to standard-dose CT. Conclusion: The optimized CycleGAN technique may allow the synthesis of diagnostically acceptable images from ULDCT of the chest for the evaluation of pulmonary tuberculosis.

• Journal of the Korean Association of Geographic Information Studies
• /
• v.27 no.2
• /
• pp.34-44
• /
• 2024

Recently, due to abnormal climate caused by climate change, natural disasters such as floods, landslides, and soil outflows are rapidly increasing. In Korea, more than 63% of the land is vulnerable to slope disasters due to the geographical characteristics of mountainous areas, and in particular, Quarry mines soil and rocks, so there is a high risk of landslides not only inside the workplace but also outside.Accordingly, this study built a DEM using UAV and aviation LiDAR for monitoring the quarry, conducted a time series change analysis, and proposed an optimal DEM construction method for monitoring the soil collection site. For DEM construction, UAV and LiDAR-based Point Cloud were built, and the ground was extracted using three algorithms: Aggressive Classification (AC), Conservative Classification (CC), and Standard Classification (SC). UAV and LiDAR-based DEM constructed according to the algorithm evaluated accuracy through comparison with digital map-based DEM.

• Journal of Korean Society of Coastal and Ocean Engineers
• /
• v.32 no.2
• /
• pp.135-145
• /
• 2020

Wave observation is widely used to direct observation method for observing the water surface elevation using wave buoy or pressure gauge and remote-sensing wave observation method. The wave buoy and pressure gauge can produce high-quality wave data but have disadvantages of the high risk of damage and loss of the instrument, and high maintenance cost in the offshore area. On the other hand, remote observation method such as radar is easy to maintain by installing the equipment on the land, but the accuracy is somewhat lower than the direct observation method. This study investigates the data quality of MIROS Wave and Current Radar (MWR) installed at Dokdo and improve the data quality of remote wave observation data using the wave buoy (CWB) observation data operated by the Korea Meteorological Administration. We applied and developed the three types of wave data quality control; 1) the combined use (Optimal Filter) of the filter designed by MIROS (Reduce Noise Frequency, Phillips Check, Energy Level Check), 2) Spike Test Algorithm (Spike Test) developed by OOI (Ocean Observatories Initiative) and 3) a new filter (H-Ts QC) using the significant wave height-period relationship. As a result, the wave observation data of MWR using three quality control have some reliability about the significant wave height. On the other hand, there are still some errors in the significant wave period, so improvements are required. Also, since the wave observation data of MWR is different somewhat from the CWB data in high waves of over 3 m, further research such as collection and analysis of long-term remote wave observation data and filter development is necessary.

• Korean Journal of Remote Sensing
• /
• v.37 no.5_1
• /
• pp.989-998
• /
• 2021

In the utilization of optical satellite imagery, which is greatly affected by clouds, periodic composite technique is a useful method to minimize the influence of clouds. Recently, a technique for selecting the optimal pixel that is least affected by the cloud and shadow during a certain period by directly inputting cloud and cloud shadow information during period compositing has been proposed. Accurate extraction of clouds and cloud shadowsis essential in order to derive optimal composite results. Also, in the case of an surface targets where spectral information is important, such as crops, the loss of spectral information should be minimized during cloud-free compositing. In thisstudy, clouds using two spectral indicators (Haze Optimized Tranformation and MeanVis) were used to derive a detection technique with low loss ofspectral information while maintaining high detection accuracy of clouds and cloud shadowsfor cabbage fieldsin the highlands of Gangwon-do. These detection results were compared and analyzed with cloud and cloud shadow information provided by Sentinel-2A/B. As a result of analyzing data from 2019 to 2021, cloud information from Sentinel-2A/B satellites showed detection accuracy with an F1 value of 0.91, but bright artifacts were falsely detected as clouds. On the other hand, the cloud detection result obtained by applying the threshold (=0.05) to the HOT showed relatively low detection accuracy (F1=0.72), but the loss ofspectral information was minimized due to the small number of false positives. In the case of cloud shadows, only minimal shadows were detected in the Sentinel-2A/B additional layer, but when a threshold (= 0.015) was applied to MeanVis, cloud shadowsthat could be distinguished from the topographically generated shadows could be detected. By inputting spectral indicators-based cloud and shadow information,stable monthly cloud-free composited vegetation index results were obtained, and in the future, high-accuracy cloud information of Sentinel-2A/B will be input to periodic cloud-free composite for comparison.