• Journal of the Korea Society for Simulation
• /
• v.30 no.2
• /
• pp.41-48
• /
• 2021

As the rural population continues to decline and aging, the improvement of agricultural productivity is becoming more important. Early prediction of crop quality can play an important role in improving agricultural productivity and profitability. Although many researches have been conducted recently to classify diseases and predict crop yield using CNN based deep learning and transfer learning technology, there are few studies which predict postharvest crop quality early in the planting stage. In this study, a early quality prediction model is proposed for sprout ginseng, which is drawing attention as a healthy functional foods. For this end, we took pictures of ginseng seedlings in the planting stage and cultivated them through hydroponic cultivation. After harvest, quality data were labeled by classifying the quality of ginseng sprout. With this data, we build early quality prediction models using several pre-trained CNN models through transfer learning technology. And we compare the prediction performance such as learning period and accuracy between each model. The results show more than 80% prediction accuracy in all proposed models, especially ResNet152V2 based model shows the highest accuracy. Through this study, it is expected that it will be able to contribute to production and profitability by automating the existing seedling screening works, which primarily rely on manpower.

• Korean Journal of Remote Sensing
• /
• v.35 no.3
• /
• pp.457-470
• /
• 2019

Automatic Target Recognition (ATR) using Synthetic Aperture Radar (SAR) has been attracted attention in the fields of surveillance, reconnaissance, and national security due to its advantage of all-weather and day-and-night imaging capabilities. However, there have been some difficulties in automatically identifying targets in real situation due to various observational and environmental conditions. In this paper, ATR problems in Extended Operating Conditions (EOC) were investigated. In particular, we considered partial occlusions of the target (10% to 50%) and differences in the depression angle between training ($17^{\circ}$) and test data ($30^{\circ}$ and $45^{\circ}$). To simulate various occlusion conditions, SARBake algorithm was applied to Moving and Stationary Target Acquisition and Recognition (MSTAR) images. The ATR accuracies were evaluated by using the template matching and Adaboost algorithms. Experimental results on the depression angle showed that the target identification rate of the two algorithms decreased by more than 30% from the depression angle of $45^{\circ}$ to $30^{\circ}$. The accuracy of template matching was about 75.88% while Adaboost showed better results with an accuracy of about 86.80%. In the case of partial occlusion, the accuracy of template matching decreased significantly even in the slight occlusion (from 95.77% under no occlusion to 52.69% under 10% occlusion). The Adaboost algorithm showed better performance with an accuracy of 85.16% in no occlusion condition and 68.48% in 10% occlusion condition. Even in the 50% occlusion condition, the Adaboost provided an accuracy of 52.48%, which was much higher than the template matching (less than 30% under 50% occlusion).

• Korean Journal of Remote Sensing
• /
• v.38 no.6_1
• /
• pp.991-1005
• /
• 2022

Satellite imaging is an effective supplementary data source for detecting and verifying nuclear activity. It is also highly beneficial in regions with limited access and information, such as nuclear installations. Time series analysis, in particular, can identify the process of preparing for the conduction of a nuclear experiment, such as relocating equipment or changing facilities. Differences in the semantic segmentation findings of time series photos were employed in this work to detect changes in meaningful items connected to nuclear activity. Building, road, and small object datasets made of KOMPSAT 3/3A photos given by AIHub were used to train deep learning models such as U-Net, PSPNet, and Attention U-Net. To pick relevant models for targets, many model parameters were adjusted. The final change detection was carried out by including object information into the first change detection, which was obtained as the difference in semantic segmentation findings. The experiment findings demonstrated that the suggested approach could effectively identify altered pixels. Although the suggested approach is dependent on the accuracy of semantic segmentation findings, it is envisaged that as the dataset for the region of interest grows in the future, so will the relevant scope of the proposed method.

• Journal of the Korea Society of Computer and Information
• /
• v.27 no.10
• /
• pp.29-41
• /
• 2022

Recently, with the development of deep learning technology, a variety of huge models with excellent performance have been devised by pre-training massive amounts of text data. However, in order for such a model to be applied to real-life services, the inference speed must be fast and the amount of computation must be low, so the technology for model compression is attracting attention. Knowledge distillation, a representative model compression, is attracting attention as it can be used in a variety of ways as a method of transferring the knowledge already learned by the teacher model to a relatively small-sized student model. However, knowledge distillation has a limitation in that it is difficult to solve problems with low similarity to previously learned data because only knowledge necessary for solving a given problem is learned in a teacher model and knowledge distillation to a student model is performed from the same point of view. Therefore, we propose a heterogeneous knowledge distillation method in which the teacher model learns a higher-level concept rather than the knowledge required for the task that the student model needs to solve, and the teacher model distills this knowledge to the student model. In addition, through classification experiments on about 18,000 documents, we confirmed that the heterogeneous knowledge distillation method showed superior performance in all aspects of learning efficiency and accuracy compared to the traditional knowledge distillation.

• Korean Journal of Remote Sensing
• /
• v.34 no.2_2
• /
• pp.327-338
• /
• 2018

The possibility of ship detection monitoring at operational level using KOMPSAT-5 Synthetic Aperture Radar (SAR) and Automatic Identification System (AIS) data is investigated. For the analysis, the KOMPSAT-5 SLC images, which are collected from the west coast of Shinjin port and the northern coast of Jeju port are used along with portable AIS data from near the coast. The ship detection algorithm based on HVAS (Human Visual Attention System) was applied, which has significant advantages in terms of detection speed and accuracy compared to the commonly used CFAR (Constant False Alarm Rate). As a result of the integrated analysis, the ship detection from KOMPSAT-5 and AIS were generally consistent except for small vessels. Some ships detected in KOMPSAT-5 but not in AIS are due to the data absence from AIS, while it is clearly visible in KOMPSAT-5. Meanwhile, SAR imagery also has some false alarms due to ship wakes, ghost effect, and DEM error (or satellite orbit error) during object masking in land. Improving the developed ship detection algorithm and collecting reliable AIS data will contribute for building wide integrated surveillance system of marine territory at operational level.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.22 no.6
• /
• pp.19-26
• /
• 2021

The lattice structure is attracting attention from industry because of its excellent strength and stiffness, ultra-lightweight, and energy absorption capability. Despite these advantages, widespread commercialization is limited by the difficult manufacturing processes for complex shapes. Additive manufacturing is attracting attention as an optimal technology for manufacturing lattice structures as a technology capable of fabricating complex geometric shapes. In this study, a unit cell was formed using a three-dimensional coordinate method. The relative density relational equation according to the boundary box size and strut radius of the unit cell was derived. Simple cubic (SC), body-centered cubic (BCC), and face-centered cubic (FCC) with a controlled relative density were designed using modeling software. The accuracy of the equations for calculating the relative density proposed in this study secured 98.3%, 98.6%, and 96.2% reliability in SC, BCC, and FCC, respectively. A simulation of the lattice structure revealed an increase in compressive yield load with increasing relative density under the same cell arrangement condition. The compressive yield load decreased in the order of SC, BCC, and FCC under the same arrangement conditions. Finally, structural optimization for the compressive load of a 20 mm × 20 mm × 20 mm structure was possible by configuring the SC unit cells in a 3 × 3 × 3 array.

• Korean Journal of Remote Sensing
• /
• v.37 no.5_1
• /
• pp.1149-1161
• /
• 2021

With a trend of the utilization of computer vision for satellite images, cloud detection using deep learning also attracts attention recently. In this study, we conducted a U-Net cloud detection modeling using SPARCS (Spatial Procedures for Automated Removal of Cloud and Shadow) Cloud Dataset with the image data augmentation and carried out 10-fold cross-validation for an objective assessment of the model. Asthe result of the blind test for 1800 datasets with 512 by 512 pixels, relatively high performance with the accuracy of 0.821, the precision of 0.847, the recall of 0.821, the F1-score of 0.831, and the IoU (Intersection over Union) of 0.723. Although 14.5% of actual cloud shadows were misclassified as land, and 19.7% of actual clouds were misidentified as land, this can be overcome by increasing the quality and quantity of label datasets. Moreover, a state-of-the-art DeepLab V3+ model and the NAS (Neural Architecture Search) optimization technique can help the cloud detection for CAS500 (Compact Advanced Satellite 500) in South Korea.

• Structural Engineering and Mechanics
• /
• v.75 no.4
• /
• pp.507-518
• /
• 2020

With the rapid development of rail transit, rail transit noise needs to be paid more and more attention. In order to accurately and effectively analyze the characteristics of low-frequency noise, a prediction model of vibration of box girder was established based on the hybrid FE-SEA method. When the train speed is 140 km/h, 200 km/h and 250 km/h, the vibration and noise of the box girder induced by the vertical wheel-rail interaction in the frequency range of 20-500 Hz are analyzed. Detailed analysis of the energy level, sound pressure contribution, modal analysis and vibration loss power of each slab at the operating speed of 140 km /h. The results show that: (1) When the train runs at a speed of 140km/h, the roof contributes more to the sound pressure at the far sound field point. Analyzing the frequency range from 20 to 500 Hz: The top plate plays a very important role in controlling sound pressure, contributing up to 70% of the sound pressure at peak frequencies. (2) When the train is traveling at various speeds, the maximum amplitude of structural vibration and noise generated by the viaduct occurs at 50 Hz. The vibration acceleration of the box beam at the far field point and near field point is mainly concentrated in the frequency range of 31.5-100 Hz, which is consistent with the dominant frequency band of wheel-rail force. Therefore, the main frequency of reducing the vibration and noise of the box beam is 31.5-100 Hz. (3) The vibration energy level and sound pressure level of the box bridge at different speeds are basically the same. The laws of vibration energy and sound pressure follow the rules below: web

• Korean Journal of Remote Sensing
• /
• v.39 no.6_1
• /
• pp.1451-1466
• /
• 2023

In the context of maritime emergencies, the utilization of drones has rapidly increased, with a particular focus on their application in search and rescue operations. Deep learning models utilizing drone images for the rapid detection of distressed vessels and other maritime drift objects are gaining attention. However, effective training of such models necessitates a substantial amount of diverse training data that considers various weather conditions and vessel states. The lack of such data can lead to a degradation in the performance of trained models. This study aims to enhance the performance of deep learning models for distress ship detection by developing a maritime environment simulator to augment the dataset. The simulator allows for the configuration of various weather conditions, vessel states such as sinking or capsizing, and specifications and characteristics of drones and sensors. Training the deep learning model with the dataset generated through simulation resulted in improved detection performance, including accuracy and recall, when compared to models trained solely on actual drone image datasets. In particular, the accuracy of distress ship detection in adverse weather conditions, such as rain or fog, increased by approximately 2-5%, with a significant reduction in the rate of undetected instances. These results demonstrate the practical and effective contribution of the developed simulator in simulating diverse scenarios for model training. Furthermore, the distress ship detection deep learning model based on this approach is expected to be efficiently applied in maritime search and rescue operations.

• Transactions of the Korean Society of Mechanical Engineers A
• /
• v.39 no.2
• /
• pp.163-168
• /
• 2015

In order to perform estimations with high reliability, it is necessary to deal with the tail part of the cumulative distribution function (CDF) in greater detail compared to an overall CDF. The use of a generalized Pareto distribution (GPD) to model the tail part of a CDF is receiving more research attention with the goal of performing estimations with high reliability. Current studies on GPDs focus on ways to determine the appropriate number of sample points and their parameters. However, even if a proper estimation is made, it can be inaccurate as a result of an incorrect threshold value. Therefore, in this paper, a GPD based on the Akaike information criterion (AIC) is proposed to improve the accuracy of the tail model. The proposed method determines an accurate threshold value using the AIC with the overall samples before estimating the GPD over the threshold. To validate the accuracy of the method, its reliability is compared with that obtained using a general GPD model with an empirical CDF.