Earticle

As interest in beauty has increased, various studies have been conducted, and related companies have considered the anthropometric data handled between humans and interfaces as an important factor. However, owing to the nature of 3D human body scanners used to extract anthropometric data, it is difficult to accurately analyze a user's body shape until a service is provided because the user only scans and extracts data. To solve this problem, the body shape of several users was analyzed, and the collected anthropometric data were obtained using a 3D human body scanner. After processing the extracted data and the anthropometric data, a custom deep learning model was designed, the designed model was learned, and the user's body shape information was predicted to provide a service suitable for the body shape. Through this approach, it is expected that the user's body shape information can be predicted using a 3D human body scanner, based upon which a beauty service can be provide.

Activity recognition in sports is often complex task resulting from the rapid dynamic interaction within players. In this paper, pre-trained VGG-16, deep learning based hockey activity recognition model has been proposed. Own hockey dataset consisting of four main activity includes free hit, goal, penalty corner and long corner was constructed as there are no existing field hockey datasets available. Experimental results indicate that the pre-trained deep learning model generates comparative results on this challenging dataset by tweaking the hyperparameters of this pre-trained model.

In this paper, we propose a soft actor?critic (SAC) algorithm with hindsight experience replay (HER), called SACHER, which is a class of deep reinforcement learning (DRL) algorithm. SAC is an off-policy model-free DRL algorithm that outperforms earlier DRL algorithms in terms of exploration and robustness. However, in SAC, maximizing the entropy-augmented objective degrades the optimality of learning outcomes. We propose SACHER to improve the learning performance of SAC. We apply SACHER to the path planning and collision avoidance control of unmanned aerial vehicles (UAVs). We demonstrate the effectiveness of SACHER in terms of the success rate, learning speed, and collision avoidance performance of UAV operation.

We propose a novel deep neural network model for detecting human activities in untrimmed videos. The process of human activity detection in a video involves two steps: a step to extract features that are effective in recognizing human activities in a long untrimmed video, followed by a step to detect human activities from those extracted features. To extract the rich features from video segments that could express unique patterns for each activity, we employ two different convolutional neural network models, C3D and I-ResNet. For detecting human activities from the sequence of extracted feature vectors, we use BLSTM, a bi-directional recurrent neural network model. By conducting experiments with ActivityNet 200, a large-scale benchmark dataset, we show the high performance of the proposed DeepAct model.

With the advances and growth of various wireless technologies, it is imperative to implement robust Intrusion Detection Systems (IDS). This paper proposes the implementation of Deep Gated Recurrent Unit (DGRU) Based classifier as well as a wrapper-based feature extraction algorithm for Wireless IDS. We assess the performance of the DRGU IDS using the NSL-KDD benchmark dataset. Furthermore, we compare our framework to several popular algorithms including Artificial Neural Networks, Deep Long?Short Term Memory, Random Forest, Naive Bayes and Feed Forward Deep Neural Networks. The experimental outcomes demonstrate that the DGRU IDS displays a significant increase in performance over existing methods.

Due to the development and dissemination of modern technology, anyone can easily communicate using services such as social network service (SNS) through a personal computer (PC) or smartphone. The development of these technologies has caused many beneficial effects. At the same time, bad effects also occurred, one of which was the spam problem. Spam refers to unwanted or rejected information received by unspecified users. The continuous exposure of such information to service users creates inconvenience in the user's use of the service, and if filtering is not performed correctly, the quality of service deteriorates. Recently, spammers are creating more malicious spam by distorting the image of spam text so that optical character recognition (OCR)-based spam filters cannot easily detect it. Fortunately, the level of transformation of image spam circulated on social media is not serious yet. However, in the mail system, spammers (the person who sends spam) showed various modifications to the spam image for neutralizing OCR, and therefore, the same situation can happen with spam images on social media. Spammers have been shown to interfere with OCR reading through geometric transformations such as image distortion, noise addition, and blurring. Various techniques have been studied to filter image spam, but at the same time, methods of interfering with image spam identification using obfuscated images are also continuously developing. In this paper, we propose a deep learning-based spam image detection model to improve the existing OCR-based spam image detection performance and compensate for vulnerabilities. The proposed model extracts text features and image features from the image using four sub-models. First, the OCR-based text model extracts the text-related features, whether the image contains spam words, and the word embedding vector from the input image. Then, the convolution neural network-based image model extracts image obfuscation and image feature vectors from the input image. The extracted feature is determined whether it is a spam image by the final spam image classifier. As a result of evaluating the F1-score of the proposed model, the performance was about 14 points higher than the OCR-based spam image detection performance.

In a duck cage, ducks are placed in various states. In particular, if a duck is overturned and falls or dies, it will adversely affect the growing environment. In order to prevent the foregoing, it was necessary to continuously manage the cage for duck growth. This study proposes a method using an object detection algorithm to improve the foregoing. Object detection refers to the work to perform classification and localization of all objects present in the image when an input image is given. To use an object detection algorithm in a duck cage, data to be used for learning should be made and the data should be augmented to secure enough data to learn from. In addition, the time required for object detection and the accuracy of object detection are important. The study collected, processed, and augmented image data for a total of two years in 2021 and 2022 from the duck cage. Based on the objects that must be detected, the data collected as such were divided at a ratio of 9 : 1, and learning and verification were performed. The final results were visually confirmed using images different from the images used for learning. The proposed method is expected to be used for minimizing human resources in the growing process in duck cages and making the duck cages into smart farms.

As the population and income levels rise, meat consumption steadily increases annually. However, the number of farms and farmers producing meat decrease during the same period, reducing meat sufficiency. Information and Communications Technology (ICT) has begun to be applied to reduce labor and production costs of livestock farms and improve productivity. This technology can be used for rapid pregnancy diagnosis of sows; the location and size of the gestation sacs of sows are directly related to the productivity of the farm. In this study, a system proposes to determine the number of gestation sacs of sows from ultrasound images. The system used the YOLOv7-E6E model, changing the activation function from sigmoid-weighted linear unit (SiLU) to a multi-activation function (SiLU + Mish). Also, the upsampling method was modified from nearest to bicubic to improve performance. The model trained with the original model using the original data achieved mean average precision of 86.3%. When the proposed multi-activation function, upsampling, and AutoAugment were applied, the performance improved by 0.3%, 0.9%, and 0.9%, respectively. When all three proposed methods were simultaneously applied, a significant performance improvement of 3.5% to 89.8% was achieved.

Capsule endoscopy is one of the increasingly demanded diagnostic methods among patients in recent years because of its ability to observe small intestine difficulties. It is often conducted for 12 to 14 hours, but significant frames constitute only 10% of whole frames. Thus, it has been designed to automatically acquire significant frames through deep learning. For example, studies to track the position of the capsule (stomach, small intestine, etc.) or to extract lesion-related information (polyps, etc.) have been conducted. However, although grouping or labeling the training images according to similar features can improve the performance of a learning model, various attributes (such as degree of wrinkles, presence of valves, etc.) are not considered in conventional approaches. Therefore, we propose a class-labeling method that can be used to design a learning model by constructing a knowledge model focused on main lesions defined in standard terminologies for capsule endoscopy (minimal standard terminology, capsule endoscopy structured terminology). This method enables the designing of a systematic learning model by labeling detailed classes through differentiation of similar characteristics.

Fire is an extremely catastrophic disaster that leads to the destruction of forests, human assets, reduced soil fertility, land resources, and the cause of global warming. In the current decade, fire detection and its management are the major concern of several researchers to prevent social, ecological, and economic damages. To overcome such kind of losses, early fire detection, and the automatic response is very significant. Moreover, achieving high accuracy with reducing inference time and model size is also challenging for the Unmanned Aerial Vehicle (UAVs). Therefore, in this work, we enabled the VGG16 architecture for UAV in terms of reducing its learning parameters from 138 million to 11.4 million for early fire detection. The proposed system is inexpensive in terms of computation and size. The performance of our proposed work is evaluated over the custom dataset. We performed comprehensive experiments using various deep learning architectures such as VGG16, ResNet50, and the proposed CNN model. The experimental results based on the proposed model achieved an accuracy of 98% on 50 epochs.

자율 주행과 교통 감시의 가장 중요한 것은 차량 감지 기술이다. 또한, 포트홀과 같이 도로의 특정 상황 은 교통사고와 차량 파손의 원인이다. 본 논문에서는 도로의 포트홀을 자동으로 발견하기 위해서 딥 러닝 모델을 사용한다. 본 연구에서는 이미지에서 차량 및 포트홀을 감지 할 수 있도록 빠른 영역 기반 컨볼루션 신경망 (Faster R-CNN)과 인셉션 네트워크 V2 모델을 사용하여 모델을 활용하였다. 제안하는 연구를 검증하기 위해 Faster R-CNN, Single Shot Detector(SSD), YOLO 알고리즘과 성능, 정확도수, 검출 시간 및 장단점을 비교 하였다. 논문에서 제안하는 방법은 SSD 및 YOLO와 같은 기존 방법보다 좋은 성능을 보여주었다. 여기서 성능 평가의 척도는 정확도를 사용하였다. 제안된 방법은 SSD 및 YOLO와 같은 이전 방법에 비해 6%의 개선을 보여준다.

Vehicle detection is the most crucial component of automated driving and traffic monitoring. Additionally, pothole-caused bad road conditions are to blame for collisions and car damage. Deep learning models are used in the suggested work. In this study, a fast region-based convolutional neural network (Faster R-CNN) and an inception network V2 model were utilized to detect vehicles and potholes in images. To verify the proposed study, Faster R-CNN, Single Shot Detector (SSD), and YOLO algorithms were compared in performance, number of accuracy, detection time, and strengths and weaknesses. Accuracy serves as the benchmark for performance evaluation. When compared to the earlier approaches, such as SSD and YOLO, the suggested method exhibits a 6% improvement.

산사태는 인명과 사회 경제적 인프라에 심각한 피해를 초래하는 대표적인 자연재해로, 위험지역의 정밀 한 공간 분할 및 조기 탐지가 재난관리에 필수적이다. 본 연구에서는 U-Net 딥러닝 아키텍처를 기반으로 패치 크기(128×128, 256×256)에 따른 산사태 위험지역 자동 분류 성능을 비교·분석하였다. 연구 대상지는 경북 예천군 보곡면의 급경사지로, 드론 영상 및 포인트 클라우드 자료를 활용해 0.1m 해상도의 디지털 표고 모델(DEM)을 구축하고, 이를 딥러닝 학습 및 평가에 적용하였다. 분석 결과, 128×128 패치 크기 모델은 전체 정확도 70.5%의 정확도를 기록했으며, 비위험과 위험 클래스에서 F1-score 0.582, 0.599를 보여 균형 잡힌 분류 성능을 나타냈다. 반면 256×256 패치 모델은 65.96%의 정확도와 매우 낮은 0.248, 0.007의 F1-score를 기록했으며, NoData 클래스에서는 0.786의 높은 F1-score를 보여 대부분의 예측이 NoData 영역에 집중됨을 나타냈다. 이러한 결과는 작은 패치 크기가 지역적 지형 변동성을 포착하고 산사태 감지 성능을 향상시키는데 더 효과적임을 시사한다.

Landslides are natural disasters that cause severe damage to human life and socio-economic infrastructure, making precise spatial delineation and early detection of hazardous areas essential for disaster management. This study compares automated classification performance of landslide hazard zones using U-Net deep learning architecture with different patch sizes (128×128 and 256×256). The study area is a steep slope in Bogok-ri, Hyoja-myeon, Yecheon-gun, Gyeongbuk Province, where a high-resolution DEM (0.1 m) was constructed using drone imagery and point cloud data. The 128×128 patch model achieved 70.5% accuracy with F1-scores of 0.582 and 0.599 for non-hazard and hazard classes, respectively, indicating balanced performance. In contrast, the 256×256 patch model yielded 65.96% accuracy, much lower F1-scores of 0.248 and 0.007, and a higher NoData F1-score of 0.786, reflecting predictions focused on NoData zones over hazard areas. Results suggest smaller patch sizes better capture local terrain variability and enhance landslide detection performance.

As the number of vehicle drivers is increasing day by day, the risk of traffic accidents is also increasing. Among the accidents, there are tire-related accidents causes huge damage if it occurs. These kinds of accidents could be prevented through safety checks of tire, but drivers usually overlook it because they don’t have the knowledge to know what the condition of the tire and don’t want to spend time and money to safety inspection and so on. To solve these problems, we propose tire life prediction mobile application with deep-learning method to check the condition of tires simply. Also, considering the embedded environment that has low power and capacity, we apply lightweight technique called pruning

The main visual feature is surface cracks, which are caused by loopholes that are embedded into the structures due to manufacturing faults and overloading factors. The structural health monitoring must be precise and more efficient for detecting surface cracks in concrete structures. Human inspection is used to identify damage on concrete surfaces. But, these traditional visual observation techniques are not more effective for large concrete structures. Moreover, this outmoded human labor practice for crack detection is intensive, expensive, and inefficient. To predict potentially hazardous situations caused by cracks on concrete surfaces, it is crucial to have an efficient, fast, and well-organized inspection system for concrete surface cracks. The automatic crack detection system must be effective in identifying cracks, damage, and segmenting them. In this research, a deep learning (DL) algorithm model is active for crack detection in concrete structure images to assess the influence on structural health. This design work is projected to provide a fast and active solution for identifying dust/duct type to prevent power losses using an image classification model based on DL. The VGG-16 DL model significantly analyzes the precision and accuracy of identifying the crack surfaces on concrete structures. The exceptional performance of the projected model achieves a training accuracy of 99.99% and a test accuracy of 99.96%, with an F1 score of 0.9995, precision of 0.9995, and a sensitivity of 0.9997. This is more precise, costeffective, and more efficient than human resources to find the defect on concrete surfaces that may support the healthy life of concrete structures.

The rise of cryptocurrencies has also led to an increase in fraudulent activities, posing challenges in fraud detection on decentralized platforms like Ethereum. This issue is particularly pronounced in decentralized environments like Ethereum, where new transaction patterns continue to emerge. In this dynamic and changing environment, it is important to address the problem of concept drift, which refers to the continuous evolution of data patterns. To address the dynamic nature of these fraud patterns, we propose an automated hyperparameter optimization (HPO) approach using Proximal Policy Optimization (PPO). Unlike traditional HPO methods, PPO efficiently navigates the complex hyperparameter space, adapting to evolving fraud schemes with minimal human intervention. Our method enhances the adaptability and robustness of fraud detection models, effectively improving detection accuracy. Experimental results demonstrate that PPO outperforms existing HPO techniques, offering a more flexible and powerful tool for maintaining the performance of fraud detection systems in the rapidly changing cryptocurrency landscape.

Accurate classification of public policy is essential for effective policy analysis, design, comparison, and formulation across countries. However, manual classification by policy experts can lead to inconsistencies and human errors, highlighting the need for a more reliable and efficient approach. This study proposes a deep learning-based model to support policy classification using artificial intelligence. Leveraging Korean policy datasets, comprising administrative data (1988–2018), legislative data (1987–2018), and media data (1988–2020), previously curated by experts, we developed an AI model for automated policy classification based on the KoBERT language model. Designed as a supplementary tool for policy experts, this model enhances classification consistency, reduces decision-making time, and improves overall productivity. Moreover, the model enables the classification, comparison, and evaluation of diverse policies at both local and national levels, offering valuable support for strategic public policy development. The proposed model achieved a Top-1 accuracy of 62.4% and a Top-3 accuracy of 71.6%, outperforming traditional baselines and demonstrating its practical potential for real-world policy analysis.

본 연구에서는 수도권 4개 권역의 PM2.5 예보를 위해 심층신경망 모델을 개발하고, 계층별 관련성 전파를 사용하여 예 측결과에 대한 입력인자들의 기여도를 분석했다. 제안한 심층신경망 예보모델은 평균적으로 지수적중률 71%, 고농도 감 지 확률 65%, 오경보율 40% 정도의 성능을 보였다. 이러한 심층신경망 예보결과에 대한 입력인자들의 중요도를 분석하 기 위해 데이터를 지역 및 PM2.5 농도로 분류하여 계층별 관련성 전파를 수행했다. 그 결과 농도 및 지역에 상관없이 중 요도가 높은 인자는 압력, 온도, 이슬점온도, 상대습도, U, V, 이산화질소, 일산화탄소로 확인된 반면에 누적강수량, 아황 산가스, PM10, 오존은 중요도가 낮은 것으로 관찰되었다. 고농도 사례에서는 일사와 오존의 중요도가 낮아지고, 예측시 간에 근접한 시간대의 PM2.5의 중요도가 높아졌으며, CMAQ(The Community Multiscale Air Quality) 예측인자들의 중 요도도 다소 올라갔다. 지역별 입력인자들의 중요도는 대체로 유사했는데 인천과 경기북부의 경우 일사의 중요도가 낮아 지는 등 고농도 패턴의 인자중요도와 일부 비슷한 결과를 보였다. 이것은 인천과 경기북부가 다른 지역보다 고농도 데이 터가 더 자주 발생하기 때문으로 분석된다. 이러한 인자중요도 결과는 향후 고농도 적중률 향상 및 지역별 특성에 적합 한 예보모델을 개발하는데 활용하면 효과적일 것으로 기대한다.

In this study, we propose a deep neural network model for PM2.5 prediction in the four metropolitan areas, and analyze the contribution of input factors to the prediction results using layer-wise relevance propagation. The proposed deep neural network forecasting model exhibited an average exponential accuracy of 71%, a high-concentration detection probability of 65%, and a false alarm ratio of 40%. In order to analyze the importance of input factors for these deep neural network forecasting results, we classify the data into regional and PM2.5 concentration level and performed layer-wise relevance propagation. As a result, factors such as pressure, temperature, dew point temperature, relative humidity, U, V, nitrogen dioxide and carbon monoxide were found to be important, regardless of any region and concentration level. On the other hand, accumulated precipitation, sulfur dioxide, PM10, and ozone are shown to have lower importance. In highconcentration cases, the importance of solar radiation and ozone decreased, PM2.5 during time periods close to the prediction time increased, and the importance of CMAQ(The Community Multiscale Air Quality) predictive factors also showed some degree of elevation. The importance of input factors by region are generally similar; however, in the case of Incheon and Northern Gyeonggi Province, the importance of the radiation decreases, showing some similarities to the importance of factors in high concentration patterns. This is analyzed as being due to a higher occurrence of highconcentration data in Incheon and Northern Gyeonggi Province compared to other regions. These analyzing results are expected to be effective in developing forecasting models that improve the prediction accuracy of high concentration and match regional characteristics in the future.

A specific area of research in artificial intelligence, known as deep learning (DL) has turned into a strong source for the solutions of complicated issues in computer vision and many more. A real application of that is real and fake faces detection. Detection of is real and fake faces became increasingly important these days with increasing deepfake technology. Fake images pose great dangers to information security, trustworthiness in multimedia content, and even to society's stability. In his proposal the design of a deep learning-based model in VGG-16 architecture to make high accuracy and reliability distinctions between real and fake faces. The performance of the proposed model was evaluated by a number of metrics, including accuracy, specificity, recall, precision, and misclassification rate. The results showed that the model obtained an excellent accuracy of 99.61% with a very low misclassification rate of 0.39%. It obtained perfect specificity of 99.15%, which means all fake faces were identified correctly, and a precision value of 99.29%, ensuring that all faces classified as real were indeed real. The recall of the model was high, at 100%, meaning nearly all real faces were correctly identified. The obtained results are the proof of how effective DL and, in this case, using a pre-trained model like VGG-16, is at recognizing real and fake faces. It shows how strong and reliable the proposed.