• Journal of The Korean Society of Grassland and Forage Science
• /
• v.43 no.3
• /
• pp.190-198
• /
• 2023

This study was conducted to calculate the damage of Italian ryegrass (IRG) by abnormal climate using machine learning and present the damage through the map. The IRG data collected 1,384. The climate data was collected from the Korea Meteorological Administration Meteorological data open portal.The machine learning model called xDeepFM was used to detect IRG damage. The damage was calculated using climate data from the Automated Synoptic Observing System (95 sites) by machine learning. The calculation of damage was the difference between the Dry matter yield (DMY)_normal and DMY_abnormal. The normal climate was set as the 40-year of climate data according to the year of IRG data (1986~2020). The level of abnormal climate was set as a multiple of the standard deviation applying the World Meteorological Organization (WMO) standard. The DMY_normal was ranged from 5,678 to 15,188 kg/ha. The damage of IRG differed according to region and level of abnormal climate with abnormal temperature, precipitation, and wind speed from -1,380 to 1,176, -3 to 2,465, and -830 to 962 kg/ha, respectively. The maximum damage was 1,176 kg/ha when the abnormal temperature was -2 level (+1.04℃), 2,465 kg/ha when the abnormal precipitation was all level and 962 kg/ha when the abnormal wind speed was -2 level (+1.60 ㎧). The damage calculated through the WMO method was presented as an map using QGIS. There was some blank area because there was no climate data. In order to calculate the damage of blank area, it would be possible to use the automatic weather system (AWS), which provides data from more sites than the automated synoptic observing system (ASOS).

• Journal of IKEEE
• /
• v.21 no.2
• /
• pp.136-141
• /
• 2017

In this paper, we propose an algorithm for memory optimization to apply deep learning to PC. The proposed algorithm minimizes the memory and computation processing time by reducing the amount of computation processing and data required in the conventional deep learning structure in a general PC. The algorithm proposed in this paper consists of three steps: a convolution layer configuration process using a random filter with discriminating power, a data reduction process using PCA, and a CNN structure creation using SVM. The learning process is not necessary in the convolution layer construction process using the discriminating random filter, thereby shortening the learning time of the overall deep learning. PCA reduces the amount of memory and computation throughput. The creation of the CNN structure using SVM maximizes the effect of reducing the amount of memory and computational throughput required. In order to evaluate the performance of the proposed algorithm, we experimented with Yale University's Extended Yale B face database. The results show that the algorithm proposed in this paper has a similar performance recognition rate compared with the existing CNN algorithm. And it was confirmed to be excellent. Based on the algorithm proposed in this paper, it is expected that a deep learning algorithm with many data and computation processes can be implemented in a general PC.

• Proceedings of the Korean Society of Crop Science Conference
• /
• 2017.10a
• /
• pp.15-15
• /
• 2017

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.13 no.2
• /
• pp.751-770
• /
• 2019

Action recognition is an essential task in computer vision due to the variety of prospective applications, such as security surveillance, machine learning, and human-computer interaction. The availability of more video data than ever before and the lofty performance of deep convolutional neural networks also make it essential for action recognition in video. Unfortunately, limited crafted video features and the scarcity of benchmark datasets make it challenging to address the multi-person action recognition task in video data. In this work, we propose a deep convolutional neural network-based Effective Hybrid Learning (EHL) framework for two-person interaction classification in video data. Our approach exploits a pre-trained network model (the VGG16 from the University of Oxford Visual Geometry Group) and extends the Faster R-CNN (region-based convolutional neural network a state-of-the-art detector for image classification). We broaden a semi-supervised learning method combined with an active learning method to improve overall performance. Numerous types of two-person interactions exist in the real world, which makes this a challenging task. In our experiment, we consider a limited number of actions, such as hugging, fighting, linking arms, talking, and kidnapping in two environment such simple and complex. We show that our trained model with an active semi-supervised learning architecture gradually improves the performance. In a simple environment using an Intelligent Technology Laboratory (ITLab) dataset from Inha University, performance increased to 95.6% accuracy, and in a complex environment, performance reached 81% accuracy. Our method reduces data-labeling time, compared to supervised learning methods, for the ITLab dataset. We also conduct extensive experiment on Human Action Recognition benchmarks such as UT-Interaction dataset, HMDB51 dataset and obtain better performance than state-of-the-art approaches.

• Journal of Korea Multimedia Society
• /
• v.23 no.4
• /
• pp.525-538
• /
• 2020

A practical animal face classification system that classifies animals in image and video data is considered as a pivotal topic in machine learning. In this research, we are proposing a novel method of fully connected dual Deep Convolutional Neural Network (DCNN), which extracts and analyzes image features on a large scale. With the inclusion of the state of the art Batch Normalization layer and Exponential Linear Unit (ELU) layer, our proposed DCNN has gained the capability of analyzing a large amount of dataset as well as extracting more features than before. For this research, we have built our dataset containing ten thousand animal faces of ten animal classes and a dual DCNN. The significance of our network is that it has four sets of convolutional functions that work laterally with each other. We used a relatively small amount of batch size and a large number of iteration to mitigate overfitting during the training session. We have also used image augmentation to vary the shapes of the training images for the better learning process. The results demonstrate that, with an accuracy rate of 92.0%, the proposed DCNN outruns its counterparts while causing less computing costs.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.21 no.10
• /
• pp.1909-1916
• /
• 2017

Recently, deep learning algorithms have good performance in various fields, but they are not actively applied to sonar systems. In this study, we carried out experiments to classify active sonar returns into a metal object such as a mine and a rock using a convolutional neural network which is one of the deep learning algorithms. Data augmentation is applied on this paper to avoid overfitting and increase performance. And we analyzed performance variation depending on hyperparameter value and change of the number of training data through data augmentation. The experiments are performed with two training data; an aspect-angle independent and an aspect-angle dependent. As a result, the performances are 88.9% and 94.9% in aspect-angle independent and dependent, respectively. These are up to 4.5% point higher than the performance obtained by applying artificial neural network and support vector machine algorithm in the previous study.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.17 no.5
• /
• pp.1377-1393
• /
• 2023

Social media is used for various purposes including entertainment, communication, information search, and voicing their thoughts and concerns about a service, product, or issue. The social media data can be used for information mining and getting insights from it. The World Health Organization has listed COVID-19 as a global epidemic since 2020. People from every aspect of life as well as the entire health system have been severely impacted by this pandemic. Even now, after almost three years of the pandemic declaration, the fear caused by the COVID-19 virus leading to higher depression, stress, and anxiety levels has not been fully overcome. This has also triggered numerous kinds of discussions covering various aspects of the pandemic on the social media platforms. Among these aspects is the part focused on vaccines developed by different countries, their features and the advantages and disadvantages associated with each vaccine. Social media users often share their thoughts about vaccinations and vaccines. This data can be used to determine the popularity levels of vaccines, which can provide the producers with some insight for future decision making about their product. In this article, we used Twitter data for the vaccine popularity detection. We gathered data by scraping tweets about various vaccines from different countries. After that, various machine learning and deep learning models, i.e., naive bayes, decision tree, support vector machines, k-nearest neighbor, and deep neural network are used for sentiment analysis to determine the popularity of each vaccine. The results of experiments show that the proposed deep neural network model outperforms the other models by achieving 97.87% accuracy.

• Journal of Broadcast Engineering
• /
• v.25 no.7
• /
• pp.1081-1094
• /
• 2020

With the recent development of deep learning, most computer vision-related tasks are being solved with deep learning-based network technologies such as CNN and RNN. Computer vision tasks such as object detection or object segmentation use intermediate features extracted from the same backbone such as Resnet or FPN for training and inference for object detection and segmentation. In this paper, an experiment was conducted to find out the compression efficiency and the effect of encoding on task inference performance when the features extracted in the intermediate stage of CNN are encoded. The feature map that combines the features of 256 channels into one image and the original image were encoded in HEVC to compare and analyze the inference performance for object detection and segmentation. Since the intermediate feature map encodes the five levels of feature maps (P2 to P6), the image size and resolution are increased compared to the original image. However, when the degree of compression is weakened, the use of feature maps yields similar or better inference results to the inference performance of the original image.

• Journal of the Korean Society of Industry Convergence
• /
• v.25 no.3
• /
• pp.467-476
• /
• 2022

The Bank of Korea raised the benchmark interest rate by a quarter percentage point to 1.75 percent per year, and analysts predict that South Korea's policy rate will reach 2.00 percent by the end of calendar year 2022. Furthermore, because market volatility has been significantly increased by a variety of factors, including rising rates, inflation, and market volatility, many investors have struggled to meet their financial objectives or deliver returns. Banks and financial institutions are attempting to provide Robo-Advisors to manage client portfolios without human intervention in this situation. In this regard, determining the best hyper-parameter combination is becoming increasingly important. This study compares some activation functions of the Deep Deterministic Policy Gradient(DDPG) and Twin-delayed Deep Deterministic Policy Gradient (TD3) Algorithms to choose a sequence of actions that maximizes long-term reward. The DDPG and TD3 outperformed its benchmark index, according to the results. One reason for this is that we need to understand the action probabilities in order to choose an action and receive a reward, which we then compare to the state value to determine an advantage. As interest in machine learning has grown and research into deep reinforcement learning has become more active, finding an optimal hyper-parameter combination for DDPG and TD3 has become increasingly important.

• Progress in Medical Physics
• /
• v.32 no.4
• /
• pp.172-178
• /
• 2021

Purpose: This study aimed to develop a deep learning architecture combining two task models to generate synthetic computed tomography (sCT) images from low-tesla magnetic resonance (MR) images to improve metallic marker visibility. Methods: Twenty-three patients with cervical cancer treated with intracavitary radiotherapy (ICR) were retrospectively enrolled, and images were acquired using both a computed tomography (CT) scanner and a low-tesla MR machine. The CT images were aligned to the corresponding MR images using a deformable registration, and the metallic dummy source markers were delineated using threshold-based segmentation followed by manual modification. The deformed CT (dCT), MR, and segmentation mask pairs were used for training and testing. The sCT generation model has a cascaded three-dimensional (3D) U-Net-based architecture that converts MR images to CT images and segments the metallic marker. The performance of the model was evaluated with intensity-based comparison metrics. Results: The proposed model with segmentation loss outperformed the 3D U-Net in terms of errors between the sCT and dCT. The structural similarity score difference was not significant. Conclusions: Our study shows the two-task-based deep learning models for generating the sCT images using low-tesla MR images for 3D ICR. This approach will be useful to the MR-only workflow in high-dose-rate brachytherapy.