Earticle

A system that recognizes unusual situations on the road in real-time and alerts them to the nearby vehicles helps ensure smooth and safe driving. In this paper we propose a method to accurately recognize abnormal road situations by combining object detection with hard negative mining and image classification based on YOLOV8 model. By combining object detection and image classification models, we gain the advantage of not only detecting abnormal road situations but also understanding their scale.

The battery market is experiencing rapid growth due to advancements in technology and increased recycling efforts. Verifying the suitability of developed batteries through rate capability experiments, which measure capacity based on charging and discharging speeds, is essential but resource-intensive and time-consuming. This research proposes a method to predict battery rate capability using a polynomial regression model based on similar data groups, aiming to shorten these experiments. The research was conducted in two main stages, namely the construction of the dataset and the development of the predictive model. Data was collected from experimental graphs in existing literature and new experiments on Coin Cell batteries. Through preprocessing steps including deduplication, interpolation, and extrapolation, a comprehensive dataset was created. A combined Quadratic and Linear Piecewise Interpolation method was developed to handle missing data efficiently. In the model development stage, polynomial regression models were created for groups of similar battery data, allowing accurate predictions for partial rate capability experiments. Experimental results demonstrated high accuracy, significantly reducing the need for extensive testing. The proposed method offers substantial time and resource savings, enhancing the efficiency of the battery development process.

The development of the Internet of Things (IoT) has created new issues in network security due to the diverse resource-constrained nature of IoT devices and the massive volume of generated heterogeneous data. One of the most important procedures in network security is intrusion detection system (IDS). The goal of intrusion detection is to locate and stop harmful activity within the network. Machine learning methods are employed to create precise IDS models. This article provides a practical overview of tree-based Machine Learning (ML) algorithms for intrusion detection. It delves into the application of Random Forest (RF), Decision Tree (DT), AdaBoost, and the J48 classifier in the context of network traffic security. The NSL-KDD data set is used to evaluate these approaches. According to experimental results, the Random Forest Classifier outperforms the other techniques.

Recent advancements in Deep Learning (DL) techniques for fire detection have mitigated various ecological, economic, and environmental impacts. However, single models in existing literature often perform poorly due to their inability to capture relevant features with limited data. In this study, we therefore develop an innovative framework for effective fire detection using teacher-student collaborative knowledge distillation. The framework comprises two main components: the teacher model, which leverages a pretrained InceptionV3, and a customized student model designed to inherit the knowledge from the pretrained InceptionV3 to a pruned student model. The proposed network is evaluated and compared against several competitive techniques using two datasets, MAFire and Yar, with different evaluation metrics, offering higher performance and lower computational cost.

The ASV-SR method introduces an innovative approach to single-image super-resolution (SISR) by integrating adaptive Stochastic Variation within a diffusion model. This combination effectively captures pixel interactions and various patterns, addressing long-range dependencies in images and overcoming the limitations of traditional deterministic SISR methods. Extensive evaluations on diverse image datasets, including PSNR, SSIM, and LPIPS metrics, reveal that the proposed model outperforms current state-of-the-art techniques. Additionally, the incorporation of a modified SWIN transformer (MST) enhances feature extraction, improving the model's adaptability and efficiency in tackling SISR challenges. This comprehensive approach underscores the significance of incorporating stochastic processes like stochastic variation to advance image super-resolution.

The rapid advancement of imaging technology has led to a surge in the volume of image data, making it challenging for image owners to efficiently store and process them. To address these issues, many are turning to cloud service providers (CSP) for their powerful storage and computational resources. Despite this convenience, reliance on cloud servers to enable computationally demanding computer vision applications such as content-based image retrieval (CBIR), poses significant privacy risks. As images may contain personally identifiable information and they may be subjected to copyright. In this regard, a straightforward solution is to encrypt images on the users’ end before sharing them with the third-party owned servers. However, the main challenge is to find a better trade-off between privacy and data usability in a cost-effective manner. Therefore, this paper presents a privacy preserving CBIR scheme that leverages the recent advancements of incorporating sub-block processing in perceptual encryption (PE) for enhanced security. In addition, our image retrieval scheme is histogram based that combines color and edge information with discrete wavelet transform; therefore, it is invariant to the encryption transformation functions. The simulation results show that our privacy preserving CBIR achieves the same retrieval performance as that of the plain images while delivering better security than the conventional privacy preserving CBIR techniques.

The identification of key feature selection plays a significant role in accurate population projection, which is an essential aspect of demographic statistics. The goal of this paper is to investigate the importance of the different features in population projection by using four advanced feature analysis techniques i.e. Canonical Correlation Analysis (CCA), Linear Discriminant Analysis (LDA) Fast Independent Component Analysis (FICA) and Principal Component Analysis (PCA). This analysis is important to determine the major factors that affect population change. The identification and ranking of these predictors can enhance demographic forecasting and policy planning. We utilized Koran population data from the UN Population Division dataset and evaluated the above four methods. The experimental results reveal that LDA achieved the lowest performance in selecting the most appropriate features, while PCA is the most efficient in selecting an effective feature with the highest variance. These insights build up the knowledge of population change and refine the projection models.

Recent advancements in data-driven methodologies have brought significant attention to the computational prediction of material properties. Traditional machine learning (ML) approaches have struggled to achieve high accuracy due to the complex relationships between a material's structure and its properties. To address this challenge, in this work, we present an ML framework for predicting the stability of silicon (Si) and Si-based alkaline metal alloys with reduced error. This emphasizes the model transferability to discover new silicon alloys with diverse electronic configurations and structures. We explore the effectiveness of two atomic structural descriptors including X-ray diffraction (XRD) and sine coulomb matrix (SCM). The dynamic ensemble learning (DEL) model is trained and evaluated using 750 Si alloys from the materials project database (MPD) and optimized via ensemble learning. The results indicate that the XRD descriptor with DEL performs most reliably for formation energy, total energy and packaging fraction prediction, showing the model robustness and transferability for ultimate efficient silicon anode’s material synthesis.

Recognizing anomalies in surveillance is crucial for public safety to identify events that deviate from normal patterns. Visual information is essential for effective anomaly recognition; however, audio data can enhance recognition accuracy by providing additional context. Despite this, existing systems only utilize visual information, overlooking the potential of audio modalities in anomaly recognition. This paper introduces a multi-modal framework for anomaly recognition through active learning, integrating audio and visual modalities to enhance anomaly prediction. The framework extracts features using a pretrained ResNet-50 convolutional neural network (CNN) model from the visual and audio data. The extracted features are then forwarded to the Bi-Directional Long Short-Term Memory (Bi-LSTM) network for temporal feature learning. These features are then fused and fed into a classification layer for final prediction. The proposed framework's performance is assessed on a benchmark dataset and yields promising results.

As the number of single-person households increases in South Korea, there is a growing demand for more personalized and space-efficient interior design, particularly among the MZ generation who value individuality. Recently, AI-powered services are being developed for efficient interior design. These services utilize indoor photographs to create digital twin-based 3D interior design programs. However, the quality of service varies significantly depending on the algorithm used. In response to this challenge, this study compares and analyzes the image segmentation performance of Grounded SAM and FastSAM, both derived from the Segment Anything Model (SAM) announced by Meta in early 2023. The ADE20K dataset, related to interior design, and the DAVIS2016 dataset, which focuses on single-object segmentation, were used to evaluate the accuracy and processing speed of the two models and to explore their applicability in real-world interior design workflows. The experimental results shows that Grounded SAM outperforms FastSAM in terms of object recognition accuracy. This research will offer valuable criteria for model selection in the automation of interior design and the development of AR/VR applications.

In this study, we analyzed animal registration data to identify the most popular dog breeds raised in South Korea. And then, a dataset was collected for the identified dog breeds and used to perform transfer learning on the YOLOv8 model to develop a breed classification model, and the classification accuracy was measured for each dog breed. The accuracy of classifying dog breeds by breed was confirmed to be at least 84% and up to 100%.

Recent advancements in object detection increasingly leverage end-to-end transformer architectures. However, many studies in this domain have applied transformer structures, originally designed for natural language processing, directly to object detection models. This direct application can lead to issues such as skipping self-attention in first decoder layer and the prediction of duplicate objects during training. In this study, we propose a novel approach to address these challenges by reversing the attention order in the transformer decoder from the self-cross to a cross-self structure. This modification structurally prevents the initial attention skip and mitigates the issue of predicting the same object multiple times by delaying the implementation of self-attention. Experimental results demonstrate that reversing the attention order in the decoder improves both the training loss and test performance across all stages of the learning process.

Anomaly detection systems for Industrial Control System (ICS) cybersecurity are designed to identify irregularities in network packets or operational data. However, they cannot detect attacks like Stuxnet, which physically injects malicious control logic. While existing studies on control logic modulation address this issue, they rely on separate storage and produce false positives. To overcome these limitations, this paper proposes an anomaly detection method that embeds PLC control logic, preserving its structure. By training the model on this embedded control logic, it learns to detect anomalies effectively. Experiments using the PLC control logic from a power plant's water treatment system confirmed that the proposed method successfully detects anomalous control logic.

The Industrial Control System (ICS) is an environment composed of control systems and field devices used to automate industrial processes. With the technological development of the 4th Industrial Revolution, as the connection between the ICS and the external network expands, the risk of exposing cyber threats to the facilities is increasing. Accordingly, cyber-attack response exercises are being implemented around the world using cyber-attack scenarios. A lot of studies are being conducted on evaluating cyber-attack response exercises as well. However, most studies focus only on evaluating the performance of response activities rather than evaluating the cyber-attack scenarios used for these exercises. Our research presents a quantitative evaluation framework to evaluate the quality of cyber-attack scenarios used in cyber-attack response exercises. Our proposed framework consists of a total of 2 stages and determines whether to use the cyber-attack scenario for cyberattack response exercise.

This paper proposes an optimization technique for voxel ambient occlusion through adaptive cone tracing. The voxel ambient occlusion technique involves voxelizing a mesh scene, storing the reflected light on the voxels, and then projecting cones in the direction of the surface normal to sample the stored reflected light, thereby expressing shading due to ambient occlusion. When applying this to volume data, where surfaces are often ambiguous, it is necessary to define a threshold density value that assumes a surface is formed where reflection occurs based on the volume data's density distribution, making the technique sensitive to noise generated during image acquisition. As a solution, this study confirms that by calculating the distance from the bounding box during cone tracing and adjusting the number and aperture angle of the cones based on this distance, it is possible to maintain realistic shadow representation in critical areas while reducing computational load and improving speed.

User authentication is a key element of security systems, requiring technologies that enhance efficiency and reliability. Although traditional fingerprint recognition is highly reliable, it requires user participation for authentication, which reduces its efficiency. To address this issue, non-intrusive and highly reliable biometric technologies, such as iris recognition, are gaining attention. In this paper, we propose a wristwatchtype biometric authentication system that utilizes electromyogram (EMG) signals, which are easy to implement in wearable systems, along with artificial intelligence (AI) hardware accelerator technology. To achieve this, a fieldprogrammable gate array (FPGA)-based hardware accelerator was utilized, with the Python on Zynq (PYNQ) platform specifically employed to maximize parallel processing capabilities and enhance the performance of the user authentication system. EMG signals were acquired through a wristwatch-type EMG sensor with two channels, and signal processing was conducted using the empirical mode decomposition (EMD) method. The artificial intelligence network employed a convolutional neural network (CNN)-long short-term memory (LSTM) architecture. This approach achieved 98.7% accuracy and a 0.5 ms response time for user authentication across four users.

With the advancement of deep learning technology, research in time series forecasting is thriving across various fields. In the financial sector, where time series data is complex and volatile, making accurate predictions challenging, the importance of such research is growing as more people invest in financial markets. While deep learning models such as Autoencoders, Recurrent Neural Networks, Long Short-Term Memory networks, and Gated Recurrent Units are actively used in financial forecasting, the Transformer model, known for its efficiency and ability to address long-term dependency issues, has predominantly been applied to stock prediction through market sentiment analysis based on textual information rather than technical price predictions. Moreover, while there is extensive research on stock forecasting, there is a notable lack of studies on Exchange Traded Funds. This study aims to bridge this gap by using Transformer models to forecast future on Exchange Traded Funds prices and performing a comparative analysis of Transformer models of various sizes.

Oral and maxillofacial reconstructive surgery demands high levels of precision and expertise, requiring extensive practice. However, repeated practice is often constrained by various real-world limitations. This paper proposes the development of a simulator that leverages virtual reality technology to provide an immersive experience in oral and maxillofacial reconstructive surgery, reducing these constraints. The simulator replicates each step of the surgical procedure, creates 3D models of the reconstruction area, and implements interaction methods necessary for performing the surgery. The proposed solution utilizes virtual reality equipment, including a head-mounted display and controllers, enabling users to experience the surgery and generate 3D model guides applicable to reconstructive surgery. Future work will involve user evaluations and accuracy assessments of the generated 3D models to validate the effectiveness of the proposed approach.

This paper presents an innovative method for reconstructing 3D models from 2D stereoscopic images captured from the front, back, left, and right sides of an object at 90 - degree horizontal rotations. The proposed process involves several key steps: plant class segmentation using artificial intelligence (AI), disparity and depth mapping from stereo images, point cloud generation, merging multiple point clouds into a single unified point cloud, and mesh application to ensure surface continuity of the 3D model. Recognizing the critical role of accurate segmentation in 3D reconstruction, this study compares two AI segmentation architectures-YOLOv8 and Detectron2-to determine which performs better in terms of segmentation accuracy, training speed, and memory consumption for the plant class. This research focuses on the 3D reconstruction phase of a parent study titled "Investigating Deep Learning for Predicting and Simulating Plant Growth Structures: A Preparatory Effort Towards the Digital Twin Paradigm in Agriculture," whose dataset comprises 2D stereo images of plants that require 3D visualization. Its long-term vision is to develop a user experience that assists farmers in making informed decisions by leveraging predictive models and 3D visualizations of crop growth.

Sugarcane is an essential crop in the global agriculture industry. There are lot of diseases in plants of growing sugarcane typically involve in five classes. These diseases consist of Mosaic, Red rot, Yellow, Rust and Healthy. Therefore, this study used to train and testi a deep learning model comprising of 2521 Sugar cane image dataset of disease-infected leaves. This research provides a sequential model for the classification of sugar cane using convolutional neural network. This study used sequential network in which ten layers are adjusted for the classification of these Mosaic, Red rot, yellow, Rust and healthy diseases. The accuracy of the proposed method works better in comparison with the previously used techniques.

The integration of technology into agriculture crop recommendation and Prediction has significantly transformed local and global agricultural productivity. Machine learning, has played a crucial role in refining this technology, offering substantial benefits to farmers, especially those operating on a small-scale farming. By using various algorithms, these technological tools have become highly effective, enabling precise predictions with minimal deviation in expected crop growth. This research highlights how different machine learning models, typically used individually, can be integrated to enhance device programming. The study underscores the impact of information technology on agriculture, demonstrating how ensemble algorithms can empower the industry to consistently achieve targeted production levels.

Maize is known as one of the healthiest diets in the world, but its productivity is critically harmed by various diseases, with blight, common rust, and gray leaf spot being the most common. Early and accurate detection of these diseases is challenging. We have developed a CNN-based Sequential Model for disease classification, which aids farmers in applying appropriate treatments. Although maize is a vital global staple, its productivity is often threatened by viral leaf diseases, leading to substantial yield losses. Timely and accurate detection of these diseases is essential for effective crop management. This study introduces a deep neural network (DNN) designed to identify maize leaf diseases—specifically Blight, Gray Leaf Spot, and Common Rust—by extracting complex image features. An attention mechanism helps the model focus on critical image areas, enhancing interpretability and robustness. Validation experiments demonstrate the model's efficiency, confirming its potential as a reliable tool for precision agriculture.

The rise of cloud-native infrastructures has accelerated the adoption of OpenTelemetry as the leading standard for monitoring system performance. OpenTelemetry's well-established utility in software observability gives way to significant limitations when applied to hardware and real-time systems. This paper explores the challenges of integrating OpenTelemetry with hybrid cloudhardware environments, drawing on real-world use cases such as the Israel-Lebanon smart device sabotage. By highlighting performance deficiencies, protocol incompatibilities, and security vulnerabilities, we illustrate how OpenTelemetry's design shortcomings diminish its utility in hardware-intensive applications. We offer recommendations for improving device compatibility and mitigating cybersecurity threats. The document seeks to provide an exhaustive analysis and a strategy for enhancing telemetry in hybrid settings.

The enactment of Republic Act 11106 establishes Filipino Sign Language (FSL) as the primary mode of communication within the deaf-mute community in the Philippines. However, this legal recognition has highlighted a significant communication gap between the deaf-mute and non-deaf-mute populations, as the latter typically does not understand FSL. This study introduces a mobile application designed to bridge this gap by translating FSL gestures into textual sentences. The application leverages a CNN-BiLSTM deep learning architecture integrated with Mistral 7B, a state-of-the-art Large Language Model (LLM), to recognize continuous multi-sign gestures and translate them into coherent text. To evaluate the system’s effectiveness, two gesture recognition models were compared based on Word Error Rate (WER), calculated using the Levenshtein distance to measure word-level discrepancies. The 1080p30 model with a stride of 5 and a window size of 30 frames achieved a WER of 27.02%, while the 720p60 model achieved, with a stride of 5 and a window size of 60 frames, a WER of 43.37%. The superior performance of the 1080p30 model is attributed to its higher spatial resolution. This research addresses the critical need for accessible communication tools, offering a solution that enhances inclusivity for the Filipino deaf community.

This paper introduces a portable flexible wrist rehabilitation sensor system designed to support patients with hand function impairments from stroke or injury. The system collects flexible sensor data before and after rehabilitation to train a specialized AI model, utilizing a Long Short-Term Memory (LSTM) network for real-time analysis. This model evaluates wrist rehabilitation performance and degrees of freedom, using embedded sensors to classify and predict hand movements. Additionally, a user-friendly GUI allows patients to monitor their recovery progress. Compared to traditional rigid exoskeletons, this flexible sensor system offers a comfortable, natural hand simulation, reduces costs, and enhances rehabilitation customization and market viability.

In digital forensics, timestamps are crucial for analyzing key events in chronological order, making them vital to forensic investigations. This paper presents a method for detecting attempts to conceal a crime through the manipulation of timestamps on IT devices. Specifically, we introduce a technique to identify timestamp manipulation in a car’s infotainment system when the suspect's smartphone is connected via Bluetooth. The method involves extracting and analyzing Bluetooth Host Controller Interface (HCI) logs, Bluetooth logs, and system logs from both the smartphone and the infotainment system. Scenario-based experiments demonstrate the effectiveness of this technique in detecting timestamp manipulation on the infotainment system.