Earticle

In recent years, machine learning algorithms are continuously being used and expanded in various fields, such as facial recognition, signal processing, personal authentication, and stock prediction. In particular, various algorithms, such as deep learning, reinforcement learning, and Q-learning, are continuously being improved. Among these algorithms, the expansion of deep learning is rapidly changing. Nevertheless, machine learning algorithms have not yet been applied in several fields, such as personal authentication technology. This technology is an essential tool in the digital information era, walking recognition technology as promising biometrics, and technology for solving state-space problems. Therefore, algorithm technologies of deep learning, reinforcement learning, and Q-learning, which are typical machine learning algorithms in various fields, such as agricultural technology, personal authentication, wireless network, game, biometric recognition, and image recognition, are being improved and expanded in this paper.

Railway transportation is the main land-based transportation in most countries. Accordingly, railway-transportation safety has always been a key issue for many researchers. Railway pedestrian accidents are the main reasons of railway-transportation casualties. In this study, we conduct experiments to determine which of the latest convolutional neural network models and algorithms are appropriate to build pedestrian railroad accident prevention systems. When a drone cruises over a pre-specified path and altitude, the real-time status around the rail is recorded, following which the image information is transmitted back to the server in time. Subsequently, the images are analyzed to determine whether pedestrians are present around the railroads, and a speed-deceleration order is immediately sent to the train driver, resulting in a reduction of the instances of pedestrian railroad accidents. This is the first part of an envisioned drone-based intelligent security system. This system can effectively address the problem of insufficient manual police force.

Diabetes is a metabolic disease affecting a multitude of people worldwide. Its incidence rates are increasing alarmingly every year. If untreated, diabetes-related complications in many vital organs of the body may turn fatal. Early detection of diabetes is very important for timely treatment which can stop the disease progressing to such complications. RR-interval signals known as heart rate variability (HRV) signals (derived from electrocardiogram (ECG) signals) can be effectively used for the non-invasive detection of diabetes. This research paper presents a methodology for classification of diabetic and normal HRV signals using deep learning architectures. We employ long short-term memory (LSTM), convolutional neural network (CNN) and its combinations for extracting complex temporal dynamic features of the input HRV data. These features are passed into support vector machine (SVM) for classification. We have obtained the performance improvement of 0.03% and 0.06% in CNN and CNN-LSTM architecture respectively compared to our earlier work without using SVM. The classification system proposed can help the clinicians to diagnose diabetes using ECG signals with a very high accuracy of 95.7%.

We developed a deep learning-based algorithm with plant fruit images to predict the quantitative traits, fruit size, and weight. Highbush blueberry was selected as a model plant because of its commercial importance. Mask R-CNN was adopted for a deep learning guidance model to predict fruits' width, length, and weight. The deep learning algorithm had a high performance on object detection and image segmentation with more than 90% accuracy and detection rate.

In this research, we have explored state-of-the-art deep learning algorithms for face emotion recognition. In this regard, we utilize a transfer learning technique for recognizing seven classes of emotions using image classifiers such as MobileNetv2, GoogleNet, ResNet101, VGGNet19. In addition, we also compared our results with a specially designed deep learning model for face emotion recognition such as “Deep Emotion”. We have utilized the dataset FER2013 exploited by many researchers. We have analyzed the accuracy, time of training, complexity in terms of layers, and other parameters. After training on FER2013, we have deployed each model on a Korean dataset obtained from Korean dramas videos containing images of celebrities. Our ultimate goal was to label all unknown Korean dataset after training on FER2013. We have presented three different programs, i.e. (i) automatic labeling, (ii) transfer learning-based models for unknown images, (iii) deep-emotion model for face emotion recognition. We have presented a detailed analysis of all models and our corresponding programs.

미세먼지에 대한 심각성이 사회적으로 대두됨에 따라 대중들은 미세먼지 예보에 대한 정보의 높은 신뢰성을 요구하고 있다. 이에 따라 다양한 신경망 알고리즘을 이용하여 미세먼지 예측을 위한 연구가 활발히 진행되고 있다. 본 논문에서는 미세먼지 예측을 위해 다양한 알고리즘으로 연구되고 있는 신경망 알고리즘들 중 대표적인 알고리즘들의 예측 성능 비교를 진행하였다. 신경망 알고리즘 중 DNN(deep neural network), RNN(recurrent neural network), LSTM(long short-term memory)을 이용하였으며, 하이퍼 파라미터 탐색을 이용하여 최적의 예측 모델을 설계하였다. 각 모델의 예측 성능 비교 분석 결과, 실제 값과 예측 값의 변화 추이는 전반적으로 좋은 성능을 보였다. RMSE와 정확도를 기준으로 한 분석에서는 DNN 예측 모델이 다른 예측 모델에 비해 예측 오차에 대한 안정성을 갖는 것을 확인하였다.

The growing concerns on the emission of particulate matter has prompted a demand for highly reliable particulate matter forecasting. Currently, several studies on particulate matter prediction use various deep learning algorithms. In this study, we compared the predictive performances of typical neural networks used for particulate matter prediction. We used deep neural network(DNN), recurrent neural network, and long short-term memory algorithms to design an optimal predictive model on the basis of a hyperparameter search. The results of a comparative analysis of the predictive performances of the models indicate that the variation trend of the actual and predicted values generally showed a good performance. In the analysis based on the root mean square error and accuracy, the DNN-based prediction model showed a higher reliability for prediction errors compared with the other prediction models.

본 연구는 딥러닝 알고리즘을 이용해 어깨 초음파 영상에서 이두근 건의 정상 및 병변을 구분하고, 영상 속 삼출액 병변을 탐지하는 컴퓨터 자동 진단 성능을 평가하고자 한다. D 병원에서 진료받은 어깨 통증 환자들의 초음 파 영상 증례 260건을 사용하였다. 딥러닝으로서 구분 알고리즘에는 ResNet-50, 탐지 알고리즘에는 DeepLabV3+ 를 적용하였으며 성능 평가 지표로 ROC 곡선, AUC, F1-Score 등을 사용하였다. 결과로 구분 알고리즘에서 정확 도 95%, 정밀도 100%, 재현율 91%, AUC 94%를 탐지 알고리즘에서 전역 정확도 97%, 평균 IOU 85%, F1-Score 66% 등을 나타냈다. 본 논문의 제시 모델을 바탕으로 추가 데이터 획득 및 여러 알고리즘을 적용한다면 임상에서 초음파 자동 진단 시스템으로의 응용이 가능하다고 판단된다.

This study aims to evaluate the computer's automatic diagnosis performance to distinguish normal and lesions of biceps from shoulder ultrasound images using a deep learning algorithm and to detect exudate lesions in the images. 260 cases of ultrasound imaging of shoulder pain patients treated at D hospital were used. As deep learning, ResNet-50 was applied to the classification algorithm and DeepLabV3+ was applied to the detection algorithm, and ROC curves, AUC, and F1-Score were used as performance evaluation indicators. As a result, 95% accuracy, 100% precision, 91% reproduction rate, and 94% AUC in the classification algorithm showed 97% global accuracy, 85% average IOU, and 66% F1-Score in the detection algorithm. Based on the model presented in this paper, it is judged that the automatic ultrasound diagnosis system can be applied in clinical practice if additional data is acquired and several algorithms are applied.

이 연구는 다양한 물리검층자료를 학습한 딥러닝 알고리듬을 이용하여 저류층의 수포화도를 예측하는 대리 모델을 구축한다. 딥러닝 알고리듬으로 계산한 수포화도 추정치를 Archie 방정식 결과와비교함으로써 개발 모델의 성능 평가를 수행하였다. 이 연구는 4가지 물리검층자료(밀도, 공극률, 비저항, 감마선)를 심층신경망의 입력인자로 활용하여 수포화도를 평가하였다, 심층신경망 기법으로는 장단기메모리학습법을 사용하였으며 전형적인 다층 인공신경망과의 비교를 통해 성능을확인하였다. 장단기메모리학습법을 기반으로 수포화도를 예측한 결과, 결정계수의 값이 0.7 이상으로 우수한 성능을 보이는 것으로 확인하였다. 모델의 민감도 분석으로는 시퀀스 조정, 유정 역할전환, k-폴드 교차검증을 시행하였다. 모델의 적용 가능성은 북해 Volve 유전과 베트남 해상 유전에적용하여 성능을 검증하였다.

This study develops a surrogate model to predict water saturation from well log data using neural-network-based deep learning algorithms. The model performance is evaluated by comparing the water saturation estimates obtained using deep learning algorithms and Archie’s law. The surrogate model evaluates the water saturation of a target reservoir using four well-log data types (density, porosity, resistivity, and gamma ray). Long Short-Term Memory (LSTM) is employed as the deep neural network algorithm, and its performance is compared with that of a multi-layer artificial neural network. Prediction via the LSTM based model showed outstanding results with the coefficient of determination above 0.7. Sensitivity analysis is conducted through sequence tuning, switching of well type, and k-fold cross-validation. The applicability of the model has been validated for the Volve oilfield in the North Sea and an offshore oilfield in Vietnam.

가짜 뉴스는 정보화 시대라는 현대사회의 특성에 의해 진위 여부의 검증과는 상관없이 빠른 속도로 확대, 재생산되어 퍼진다. 전체 뉴스의 1%를 가짜라고 가정했을 경우 우리사회에 미치는 경제적 비용이 30조 원에 달한다고 하니 가짜 뉴스는 사회적, 경제적으로 매우 중요한 문제라고 할 수 있다. 이에 본 연구는 뉴스의 진위 여부를 신속하고 정확하게 확인하고자 자동화된 가짜 뉴스 탐지 모델을 개발하는데 목적을 두고 있다. 이를 위해 본 연구에서는 크롤링(crawling)을 통해 진위 여부가 밝혀진 뉴스 기사를 수집하였고, 워드 임베딩(Word2Vec, Fasttext)과 딥러닝 기법(LSTM, BiLSTM)을 이용하여 가짜 뉴스 예측 모델을 개발하였다. 실험 결과, Word2Vec과 BiLSTM의 조합이 가장 높은 84%의 정확도를 보였다.

Fake news isexpanded and reproduced rapidly regardless of their authenticity by the characteristics of modern society, called the information age. Assuming that 1% of all news are fake news, the amount of economic costs is reported to about 30 trillion Korean won. This shows that the fake news isvery important social and economic issue. Therefore, this study aims to develop an automated detection model to quickly and accurately verify the authenticity of the news. To this end, this study crawled the news data whose authenticity is verified, and developed fake news prediction models using word embedding (Word2Vec, Fasttext) and deep learning algorithms (LSTM, BiLSTM). Experimental results show that the prediction model using BiLSTM with Word2Vec achieved the best accuracy of 84%.

본 논문은 자율주행 차량 적용을 위한 객체 검출과 거리 추정을 수행하는 시스템을 제안한 다. 객체 검출은 최근 활발하게 사용되는 딥러닝 모델 YOLOv4의 특성을 이용해서 입력 이미 지 비율에 맞춰 분할 grid를 조정하고 자체 데이터셋으로 전이학습된 네트워크로 수행한다. 검 출된 객체까지의 거리는 bounding box와 homography를 이용해 추정한다. 실험 결과 제안하는 방법에서 전반적인 검출 성능 향상과 실시간에 가까운 처리 속도를 보였다. 기존 YOLOv4 대 비 전체 mAP는 4.03% 증가했다. 도심로 주행시 빈출하는 보행자, 차량 및 공사장 고깔(cone), PE드럼(drum) 등의 객체 인식 정확도가 향상되었다. 처리 속도는 약 55 FPS이다. 거리 추정 오차는 X 좌표 평균 약 5.25m, Y 좌표 평균 0.97m으로 나타났다.

This paper proposes a system that performs object detection and distance estimation for application to autonomous vehicles. Object detection is performed by a network that adjusts the split grid to the input image ratio using the characteristics of the recently actively used deep learning model YOLOv4, and is trained to a custom dataset. The distance to the detected object is estimated using a bounding box and homography. As a result of the experiment, the proposed method improved in overall detection performance and processing speed close to real-time. Compared to the existing YOLOv4, the total mAP of the proposed method increased by 4.03%. The accuracy of object recognition such as pedestrians, vehicles, construction sites, and PE drums, which frequently occur when driving to the city center, has been improved. The processing speed is approximately 55 FPS. The average of the distance estimation error was 5.25m in the X coordinate and 0.97m in the Y coordinate.

본 논문은 불완전한 데이터를 처리하기 위해 2가지의 서로 다른 기법과 이를 학습하는 알고리즘을 소개한다. 첫째방법은 손실변수가 가질 수 있는 균등한 확률로 손실값을 할당하여 불완전한 데이터를 처리하고, SVM 알고리즘으 로 이 데이터를 학습하는 것이다. 이 기법은 임의의 변수에 손실 값의 빈도가 높을수록 엔트로피가 높도록 하여 이 변수 가 결정트리에서 선택되지 않도록 하는 것이다. 이 방법은 손실 변수에 남아있는 정보를 모두 무시하고 새로운 값을 할당한다는 특징이 있다. 이에 반해 새로운 방법은 손실 값을 제외하고 남아있는 정보로 엔트로피 확률을 구하고 이를 손실 변수의 추정 값으로 사용하는 것이다. 즉, 불완전한 학습데이터로부터 소실되지 않은 많은 정보들을 이용해 소실된 일부 정보를 복구하고 딥러닝을 이용해 학습한다. 이 2가지 방법은 학습데이터에서 차례로 변수 하나를 선택하고, 이 변수에 손실된 데이터의 비율을 달리하면서 서로 다른 측정값들의 결과들과 반복적으로 비교함으로써 성능을 측정한다.

This paper introduces two different techniques for dealing with incomplete data and algorithms for learning this data. The first method is to process the incomplete data by assigning the missing value with equal probability that the missing variable can have, and learn this data with the SVM. This technique ensures that the higher the frequency of missing for any variable, the higher the entropy so that it is not selected in the decision tree. This method is characterized by ignoring all remaining information in the missing variable and assigning a new value. On the other hand, the new method is to calculate the entropy probability from the remaining information except the missing value and use it as an estimate of the missing variable. In other words, using a lot of information that is not lost from incomplete learning data to recover some missing information and learn using deep learning. These two methods measure performance by selecting one variable in turn from the training data and iteratively comparing the results of different measurements with varying proportions of data lost in the variable.

AI image generation technology has become a popular research direction in the field of AI, which is widely used in the field of digital art and conceptual design, and can also be used in the process of 3D texture mapping. This paper introduces the production process of 3D texture mapping using AI image technology, and discusses whether it can be used as a new way of 3D texture mapping to enrich the 3D texture mapping production process. Two AI deep learning models, Stable Diffusion and Midjourney, were combined to generate high-quality AI textures. Finally, the lmage to material function of substance 3D Sampler was used to convert the AI-generated textures into PBR 3D texture maps. And applied in 3D environment. This study shows that 3D texture maps generated by AI image generation technology can be used in 3D environment, which not only has short production time and high production efficiency, but also has rich changes in map styles, which can be quickly adjusted and modified according to the design scheme. However, some AI texture maps need to be manually modified before they can be used. With the continuous development of AI technology, there will be great potential for further development and innovation of AI-generated image technology in the 3D content production process in the future.

In recent years, with the breakthrough of Artificial Intelligence (AI) technology in deep learning algorithms such as Generative Adversarial Networks (GANs) and Variational Autoencoders (VAE), AI generation technology has rapidly expanded in various sub-sectors in the art field. 2022 as the explosive year of AI-generated art, especially in the creation of AI-generated art creative design, many excellent works have been born, which has improved the work efficiency of art design. This study analyzed the application design characteristics of AI generation technology in two sub fields of artistic creative design of AI painting and AI animation production , and compares the differences between traditional painting and AI painting in the field of painting. Through the research of this paper, the advantages and problems in the process of AI creative design are summarized. Although AI art designs are affected by technical limitations, there are still flaws in artworks and practical problems such as copyright and income, but it provides a strong technical guarantee in the expansion of subdivisions of artistic innovation and technology integration, and has extremely high research value.

Recognizing handwritten digits was challenging task in a couple of years ago. Thanks to machine learning algorithms, today, the issue has solved but those algorithms require much time to train and to recognize digits. Thus, using one of those algorithms to an application that works in real-time, is complex. Notwithstanding use of a trained model, if the model uses deep neural networks it requires much more time to make a prediction and becomes more complicated as well as memory usage also increases. It leads real-time application to delay and to work slowly even using trained model. A memory usage is also essential as using smaller memory of trained models works considerable faster comparing to models with huge pre-processed memory. For this work, we implemented four models on the basis of unlike algorithms which are capsule network, deep residual learning model, convolutional neural network and multinomial logistic regression to recognize handwritten digits. These models have unlike structure and they have showed a great results on MNIST before so we aim to compare them in real-time environment. The dataset MNIST seems most suitable for this work since it is popular in the field and basically used in many state-of-the-art algorithms beyond those models mentioned above. We purpose revealing most suitable algorithm to recognize handwritten digits in real-time environment. Also, we give comparisons of train and evaluation time, memory usage and other essential indexes of all four models.

Purpose Assessing the metastasis status of the sentinel lymph nodes (SLNs) for hematoxylin and eosin?stained frozen tissue sections by pathologists is an essential but tedious and time-consuming task that contributes to accurate breast cancer staging. This study aimed to review a challenge competition (HeLP 2019) for the development of automated solutions for classifying the metastasis status of breast cancer patients. Materials and Methods A total of 524 digital slides were obtained from frozen SLN sections: 297 (56.7%) from Asan Medical Center (AMC) and 227 (43.4%) from Seoul National University Bundang Hospital (SNUBH), South Korea. The slides were divided into training, development, and validation sets, where the development set comprised slides from both institutions and training and validation set included slides from only AMC and SNUBH, respectively. The algorithms were assessed for area under the receiver operating characteristic curve (AUC) and measurement of the longest metastatic tumor diameter. The final total scores were calculated as the mean of the two metrics, and the three teams with AUC values greater than 0.500 were selected for review and analysis in this study. Results The top three teams showed AUC values of 0.891, 0.809, and 0.736 and major axis prediction scores of 0.525, 0.459, and 0.387 for the validation set. The major factor that lowered the diagnostic accuracy was micro-metastasis. Conclusion In this challenge competition, accurate deep learning algorithms were developed that can be helpful for making a diagnosis on intraoperative SLN biopsy. The clinical utility of this approach was evaluated by including an external validation set from SNUBH.

본 논문에서는 딥러닝 지도학습 알고리즘을 사용한 학습 모델을 대상으로 미용 관련 피부질환 인식의 효과성을 실험적으로 비교한다. 최근 딥러닝 기술을 산업, 교육, 의료 등 다양한 분야에 적용하고 있으며, 의료 분야에서는 중요 피부질환 중 하나인 피부암 식별의 수준을 전문가 수준으로 높인 성과를 보이고 있다. 그러나 아직 피부미용과 관련된 질환에 적용한 사례가 다양하지 못하다. 따라서 딥러닝 기반 이미지 분류에 활용도가 높은 CNN 알고리즘을 비롯하여 ResNet, SE-ResNet을 적용하여 실험적으로 정확도를 비교함으로써 미용 관련 피부질환을 판단하는 효과성을 평가한다. 각 알고리즘을 적용한 학습 모델을 실험한 결과에서 CNN의 경우 평균 71.5%, ResNet은 평균 90.6%, SE-ResNet은 평균 95.3%의 정확도를 보였다. 특히 학습 깊이를 다르게하여 비교한 결과 50개의 계층 구조를 갖는 SE-ResNet-50 모델이 평균 96.2%의 정확도로 미용 관련 피부질환 식별을 위해 가장 효과적인 결과를 보였다. 본 논문의 목적은 피부 미용과 관련된 질환의 판별을 고려하여 효과적인 딥러닝 알고리즘의 학습과 방법을 연구하기 위한 것으로 이를 통해 미용 관련 피부질환 개선을 위한 서비스 개발로 확장할 수 있을 것이다.

In this paper, we empirically compare the effectiveness of training models to recognize beauty-related skin disease using supervised deep learning algorithms. Recently, deep learning algorithms are being actively applied for various fields such as industry, education, and medical. For instance, in the medical field, the ability to diagnose cutaneous cancer using deep learning based artificial intelligence has improved to the experts level. However, there are still insufficient cases applied to disease related to skin beauty. This study experimentally compares the effectiveness of identifying beauty-related skin disease by applying deep learning algorithms, considering CNN, ResNet, and SE-ResNet. The experimental results using these training models show that the accuracy of CNN is 71.5% on average, ResNet is 90.6% on average, and SE-ResNet is 95.3% on average. In particular, the SE-ResNet-50 model, which is a SE-ResNet algorithm with 50 hierarchical structures, showed the most effective result for identifying beauty-related skin diseases with an average accuracy of 96.2%. The purpose of this paper is to study effective training and methods of deep learning algorithms in consideration of the identification for beauty-related skin disease. Thus, it will be able to contribute to the development of services used to treat and easy the skin disease.

Recently, there have been many studies on machine learning. Among them, studies on reinforcement learning are actively worked. In this study, we propose a controller to control bicycle using DDPG (Deep Deterministic Policy Gradient) algorithm which is the latest deep reinforcement learning method. In this paper, we redefine the compensation function of bicycle dynamics and neural network to learn agents. When using the proposed method for data learning and control, it is possible to perform the function of not allowing the bicycle to fall over and reach the further given destination unlike the existing method. For the performance evaluation, we have experimented that the proposed algorithm works in various environments such as fixed speed, random, target point, and not determined. Finally, as a result, it is confirmed that the proposed algorithm shows better performance than the conventional neural network algorithms NAF and PPO.