As Internet of Things (IoT) technology is being used in all industries, the importance of secure and convenient firmware update technology is increasing. However, conventional FOTA (Firmware Over-The-Air) technology has a problem because the security is weak when updating firmware with a single path, and strong encryption technology cannot be utilized. Therefore, this study proposes a secure FOTA (S-FOTA) mechanism for lightweight IoT and adaptive S-FOTA ARQ (Automatic Repeat Request) mechanism. This adaptive S-FOTA ARQ mechanism considers the case where the original file cannot be recovered because of the increase in lost files due to the congested channel state and compares and analyzes the conventional method in terms of security, complexity, and transmission speed. Experimental results show that S-FOTA with 40 encrypted files reduced the attacker’s attack success rate by at least 62.58% and up to 99.99%, and S-FOTA with 40% of the total number of encrypted file segments takes at least 996.39% more time on average and up to 3374.99% more time than conventional FOTA. In addition, the transmission speed of the adaptive S-FOTA ARQ mechanism was at least 63.16% and up to 2736.36% higher than that of the conventional S-FOTA, and at least 53.89% and up to 70.89% higher than that of the conventional ARQ mechanism.

Digital watermarking is a technique for embedding information into digital content. Digital watermarking has attracted attention as a technique to combat piracy and identify artificially generated content, but it is still not robust in various situations. In this paper, we propose a frequency conversion-based hierarchical watermarking technique capable of attack detection, error correction, and owner identification. By embedding attack detection and error correction signatures in hierarchical watermarking, the proposed scheme maintains invisibility and outperforms the existing methods in capacity and robustness. We also proposed a framework to evaluate the performance of the image quality and error correction according to the type of error correction signature and the number of signature embeddings. We compared the visual quality and error correction performance of the conventional model without error correction signature and the conventional model with hamming and BCH signatures. We compared the quality by the number of signature embeddings and found that the quality deteriorates as the number of embeddings increases but is robust to attacks. By analyzing the quality and error correction ability by error correction signature type, we found that hamming codes showed better error correction performance than BCH codes and 41.31% better signature restoration performance than conventional methods.

Cats are known to express their emotions through a variety of vocalizations during interactions. These sounds reflect their emotional states, making the understanding and interpretation of these sounds crucial for more effective communication. Recent advancements in artificial intelligence has introduced research related to emotion recognition, particularly focusing on the analysis of voice data using deep learning models. Building on this background, the study aims to develop a deep learning system that classifies and generates cat sounds based on their emotional content. The classification model is trained to accurately categorize cat vocalizations by emotion. The sound generation model, which uses deep learning based models such as SampleRNN, is designed to produce cat sounds that reflect specific emotional states. The study finally proposes an integrated system that takes recorded cat vocalizations, classify them by emotion, and generate cat sounds based on user requirements.

Generative Adversarial Networks (GANs) are typically described as a two-player game between a generator and a discriminator, where the generator aims to produce realistic data, and the discriminator tries to distinguish between real and generated data. However, this setup often leads to mode collapse, where the generator produces limited variations in the data, failing to capture the full range of the target data distribution. This paper proposes a new training system to mitigate the mode collapse problem. Specifically, it extends the traditional two-player game of GANs into a multi-player game and introduces a peer-evaluation method to effectively train multiple GANs. In the peer-evaluation process, the generated samples from each GANs are evaluated by the other players. This provides external feedback, serving as an additional standard that helps GANs recognize mode failure. This cooperative yet competitive training method encourages the generators to explore and capture a broader range of the data distribution, mitigating mode collapse problem. This paper explains the detailed algorithm for peer-evaluation based multi-GANs training and validates the performance through experiments.

As the need for hygiene management increases due to COVID-19, the importance of non-contact services is gaining attention. Hands, a tool for expressing intentions and conveying information, are emerging as an alternative to computer input devices such as the keyboard and mouse. In this study, we propose a method to address public health problems that arise when using unmanned ordering machines by controlling a computer using hand gestures detected through a camera. The focus is on identifying frequently used hand gestures, especially the bending of the index finger. To this end, we develop a non-contact input device using the MediaPipe framework and the long short-term memory (LSTM) model. This approach can identify hand gestures in three-dimensional space and provides scenarios that can be applied to the fields of virtual reality (VR) and augmented reality (AR). It offers improved public health and user experience by presenting methods that can be applied to various situations such as navigation systems and unmanned ordering machines.

Advancements in artificial intelligence(AI) have significantly improved deep learning-based audio deepfake technology, which has been exploited for criminal activities. To detect audio deepfake, we propose CoNSIST, an advanced audio deepfake detection model. CoNSIST builds on AASIST, which a graph-based end-to-end model, by integrating three key components: Squeeze and Excitation, Positional Encoding, and Reformulated HS-GAL. These additions aim to enhance feature extraction, eliminate unnecessary operations, and incorporate diverse information. Our experimental results demonstrate that CoNSIST significantly outperforms existing models in detecting audio deepfakes, offering a more robust solution to combat the misuse of this technology

This study explores the differences in technical aspects and beauty between artworks generated by artificial intelligence (AI) and humans, assessing the characteristics, potential, and limitations of AI art, as well as the role of artists in depth. The results demonstrate that artworks generated by AI possess a level of technical proficiency and aesthetic value that can compete with human art, earning high appreciation among the general public. Specifically, with regard to emotional transmission and impression, AI can provide an artistic resonance comparable to that of humans; however, in artworks depicting natural landscapes, the subtle emotions and techniques of human artists surpass those of AI. This indicates that while AI can play a significant role in the field of artistic creation, AI also exhibits limitations in specific genres. The study is expected to provide deeper insights into the social acceptance and the position of AI art within the art community

Recent advancements in AI technology have led to its application across various fields. However, the lack of transparency in AI operations makes it challenging to guarantee the quality of its outputs. Therefore, we integrate requirements engineering in software engineering with conversational AI technology to ensure procedural fairness. Traditional requirements engineering research uses grammar-centered analysis, which often fails to fully interpret the semantic aspects of natural language. To solve this, we suggest combining Noam Chomsky's syntactic structure analysis with Charles Fillmore's semantic role theory. Additionally, we extend our previous research by analyzing adjectives in informal requirement sentence structures. This enables precise emotional analysis of the main characters in comics. Based on the results of the analysis, we apply the emotional states of the objects to the states in the UML state diagram. Then, we create the 3D object with Three.js based on the object that reflects the emotional states in the state diagram. With this approach, we expect to represent the emotional state of a 3D object.

The rapid advancement of artificial intelligence and machine learning technologies is driving innovation across various industries, with natural language processing offering substantial opportunities for the analysis and processing of text data. The development of effective text classification models requires several complex stages, including data exploration, preprocessing, feature extraction, model selection, hyperparameter optimization, and performance evaluation, all of which demand significant time and domain expertise. Automated machine learning (AutoML) aims to automate these processes, thus allowing practitioners without specialized knowledge to develop high-performance models efficiently. However, current AutoML frameworks are primarily designed for structured data, which presents challenges for unstructured text data, as manual intervention is often required for preprocessing and feature extraction. To address these limitations, this study proposes a web-based AutoML platform that automates text preprocessing, word embedding, model training, and evaluation. The proposed platform substantially enhances the efficiency of text classification workflows by enabling users to upload text data, automatically generate the optimal ML model, and visually present performance metrics. Experimental results across multiple text classification datasets indicate that the proposed platform achieves high levels of accuracy and precision, with particularly notable performance when utilizing a Stacked Ensemble approach. This study highlights the potential for non-experts to effectively analyze and leverage text data through automated text classification and outlines future directions to further enhance performance by integrating Large language models.

As the 4th Industrial Revolution progresses, the shipping logistics sector is becoming smarter by utilizing various core technologies such as AI, IoT, and Bigdata. In particular, the collected marine Bigdata plays a significant role in providing various services like vessel operation monitoring analysis and greenhouse gas emission evaluation, and it is also essential in shipping logistics. Although this maritime Bigdata is collected during actual vessel operations, there are instances where data is lost due to temporal and environmental factors. While It is important to identify and address the fundamental cause of such losses, it is also necessary to generate data through the utilization and analysis of the collected data. This paper develops an Emulator that repeatedly generates new location data, speed values, etc., using maritime transport data collected through empirical tests. The location data is generated by calculating the standard deviation from the collected position information, and the speed values are extracted from the generated location data. The generated data is accumulated by being inserted into the database in real-time. To demonstrate the performance of the Emulator, evperiments were conducted using 5 routes, providing its excellence.

Morphological analysis involves segmenting morphemes, the smallest units of meaning or grammatical function in a language, and assigning part-of-speech tags to each morpheme. It plays a critical role in various natural language processing tasks, such as named entity recognition and dependency parsing. Much of modern natural language processing relies on deep learning-based language models, and Korean morphological analysis can be broadly categorized into sequence-to-sequence methods and sequential labeling methods. This study proposes a morphological analysis approach using the transformer encoder for sequential labeling to perform syllable-level part-of-speech tagging, followed by morpheme restoration and tagging through a pre-analyzed dictionary. Additionally, the CBOW method was used to extract syllable-level embeddings in lower dimensions, designing a lightweight morphological analyzer model with reduced parameters. The proposed model achieves fast inference speed and low parameter usage, making it efficient for use in resource-constrained environments.

The purpose of this paper is to design and implement a wheel-driven small intelligent robot with intelligent sensor signal processing and various driving methods to overcome non-flat terrain such as slopes and steps and avoid obstacles. An eccentric gear structure was proposed to overcome non-flat terrain, optimal sensor signal processing was applied to maintain real-time balance, and an omnidirectional driving method that enables obstacle recognition and escape from a narrow space using a LiDAR sensor was proposed and designed to overcome obstacles. An optimal embedded system was designed and constructed to implement and control the intelligent elements of the robot.

decision and visual obstructions caused by various factors can lead to errors during the procedure. This paper proposes a 2D image-based real-time implant placement guiding system that predicts the implant position using 2D surgical video without the need for preoperative oral scans or 3D model generation. In the initial phase of the surgical video, two segmentation models are employed to measure prior statistics of the occlusal and incisal surfaces for each tooth. Subsequently, a single segmentation model is used to separate the occlusal and incisal surfaces, and the implant placement is predicted and guided based on the axis and length of adjacent teeth as well as the center of the prosthesis to be implanted. The system was designed and implemented using a dental phantom model, which replicates the oral structure of an actual human. The algorithm’s average execution time for guiding implant placement on 10 images was measured to be 12.14 ms, demonstrating its feasibility for real-time application in surgical video.

JavaScript is slow for high-performance image processing in web browsers and cannot directly utilize the GPU. Therefore, web plugin technology or server-based processing methods have been used. However, since web plugins are no longer supported by the latest web browsers and server processing methods become increasingly expensive as the number of users grows. In this paper, an image processing acceleration method is proposed using the latest web standards such as WASM and WebGPU in a client environment, instead of plugins or server-based methods. The final experimental results confirmed that the WASM+WebGPU-based code, which utilizes both the CPU and GPU, improved execution performance by up to 10 times compared to traditional javaScript.