The purpose of this study is to develop a computerized classification method for significant coronary artery stenosis in whole-heart coronary magnetic resonance angiography (WHCMRA) images using a three-dimensional convolutional neural network (CNN) enhanced with attention mechanisms. Our database included 951 segments from WHCMRA images (pixel size = 0.645 mm) obtained from 75 patients. Forty-two segments with significant stenosis (luminal diameter reduction ≥ 75%) were annotated on WHCMRA images by an experienced radiologist, whereas 909 segments without stenosis were annotated at representative sites. In the proposed method, high-resolution WHCMRA images (pixel size = 0.3225 mm) were generated from the original WHCMRA images using a CNN-based super-resolution model that extends conventional two-dimensional architecture into a three-dimensional framework. Volumes of interest, centered on annotated points, were extracted from these high-resolution WHCMRA images. The three-dimensional CNN for classifying coronary artery stenosis consists of two feature extractors, two attention mechanisms, and a classifier. The proposed method achieved an area under the receiver operating characteristic curve of 0.954, indicating a substantial impact on the interpretation of WHCMRA.

Underwater acoustic communication, which enables wireless data exchange in environments where radio waves are severely attenuated, plays a crucial role in enhancing the efficiency of subsea infrastructure inspection and seafloor exploration. However, underwater acoustic channels are characterized by significant delay spread, Doppler spread, and Doppler shift, all of which must be accurately estimated and compensated to ensure reliable communication. Orthogonal signal division multiplexing has proven effective in maintaining robust data transmission by jointly compensating for delay and Doppler effects. Nevertheless, correcting a large Doppler shift often requires extensive computational resources, limiting the feasibility of real-time communication. To address this challenge, we propose an optimization algorithm that adaptively restricts the Doppler search range by referencing the previously measured Doppler shift and its rate of variation. This method achieves accurate Doppler shift correction while significantly reducing computational cost. Sea trial results demonstrate that the proposed approach achieves communication performance comparable to that of the existing method while reducing the number of cross-correlation operations by 80% and the overall computation time by 62%.

Accurate recognition of cattle behavior, particularly mounting behavior, plays a crucial role in livestock management, as it aids in detecting estrus and optimizing herd productivity. Traditional behavior recognition methods are often labor- intensive and time-consuming, requiring manual intervention and reliance on predefined object categories. In this paper, we present an advanced multimodal system for black cattle mounting behavior analysis, which leverages YOLO-World, a cutting-edge object detection framework enhanced with open-vocabulary capabilities. While YOLO-World has been developed to overcome the limitations of traditional YOLO models, such as their dependence on predefined categories, our contribution lies in integrating this model into a streamlined pipeline for end-to-end detection, tracking, and action recognition of black cattle. We utilize YOLO-World's ability to perform open-vocabulary detection, allowing the system to adapt to new and unseen objects and behaviors without manual categorization. Additionally, our system is optimized for deployment on IoT edge devices, enabling real-time cattle monitoring in field conditions. This approach significantly reduces the time and effort required compared to traditional methods. Experimental results demonstrate improved accuracy in mounting behavior recognition, though challenges such as false positives persist. Overall, this work presents a scalable, efficient solution for real-time cattle behavior analysis in open-world environments, contributing to advancements in automated livestock monitoring.

In wireless power transfer (WPT) systems for electric vehicles, metallic foreign objects near the power supply unit can overheat and pose ignition risks. Detecting such objects and shutting off the power is critical. Our previous study revealed that Wi-Fi channel state information (CSI) changes significantly when metallic foreign objects pass close to the antennas. In this study, we evaluated detection accuracy using an L-shaped directional antenna setup around a designed area representing a power supply unit. The detection accuracy of foreign objects was compared between the 2.4 GHz and 5 GHz bands using two types of antenna arrangement with and without an aluminum plate simulating a vehicle. The results showed that while the differences in accuracy owing to antenna arrangement and the presence or absence of the aluminum plate were very small, the detection accuracy was lower in the 2.4 GHz band than in the 5 GHz band.

The demand for indoor monitoring systems has increased owing to the growing elderly population. Cameras are commonly used for indoor monitoring, but the field of view is limited. In this paper, we propose a human position estimation method using Wi-Fi within an observation area without complex calculations. Experiments were conducted in an 18 m² observation area inside a gymnasium, where a single access point (AP) was placed in three different positions, and 12 receiving antennas were installed around the perimeter of the observation area. Human positions were estimated on the basis of channel state information (CSI) variations, specifically the moving variance of the CSI amplitude component measured at each receiving antenna. The estimation error was evaluated by comparing the results to actual positions measured by LiDAR.

In this paper, frequency-modulated continuous wave (FMCW) radar for enabling a more comfortable traffic environment is discussed and a scheme that integrates ranging and communication is proposed. A method to increase the data rate while maintaining ranging accuracy is proposed. Data are modulated onto the sweep function of the FMCW radar. Both up-chirp and down-chirp signals are utilized for data modulation. Furthermore, multiple sweep functions are generated by applying cyclic time shifts to the sweep function. By selecting different sweep functions in accordance with the data to be modulated, a transmitted transmitting additional data bits is generated. At the receiver, the beat frequency is calculated from the reflected signal and reference signals, enabling distance measurement as in the conventional beat frequency ranging scheme. The root mean square error (RMSE) and bit error rate (BER) of the proposed scheme are evaluated through simulations and compared with those of conventional FMCW radar. The performance of the proposed scheme is better than that of the conventional scheme that modulates 1-bit data using up- and down-chirps at the same data rate. Under higher signal-to-noise ratio (SNR) conditions, the proposed scheme achieves convergence to the same RMSE as does the conventional FMCW radar without data modulation. Therefore, the proposed scheme is effective in increasing the data rate.

In this study, we investigate the optimization of wireless network selection for reliable communication in cybernetic avatar remote control. A centralized method utilizing a feedforward neural network is proposed to address the limitations of traditional methods, which rely on single metrics such as received signal strength indicator and are reactive. By integrating network-wide information and mobility data, the method predicts throughput in real time and selects the optimal network connection to ensure stable and high-quality communication. Experimental evaluations demonstrated up to 2-fold improvement in average throughput compared with the signal-based network selection method, while maintaining comparable processing times. The proposed method highlights the potential of a machine learning-based network selection method to enhance real-time QoS in dynamic environments.

In a multihop communication using low earth orbit(LEO) satellite networks between two ground stations, latency increases when access to satellites along the route is congested. In this paper, we propose a route selection method using a deep reinforcement learning algorithm to obtain a communication path with reduced delay. For each satellite, deep reinforcement learning is performed using the states of neighboring satellites to select the next hop satellite for relaying. We employ the Deep Q-Network as a reinforcement algorithm, with the queuing delay and distance to the destination ground station used as state information. The reward is defined as the difference in latency between the shortest distance route and the route selected by the algorithm. We evaluate the performance of the proposed method through computer simulations of dynamic LEO satellite constellations. The results demonstrate that the proposed method effectively avoids satellites experiencing high traffic and selects routes with shorter delays than the shortest distance route.

The rapid advancements in artificial intelligence (AI) technologies for singing voice synthesis have revolutionized music production but introduced significant challenges, including the misuse of AI-generated voices for deepfake purposes. This study proposes a framework for detecting deepfake singing voices using data augmentation techniques, the Synthetic Minority Over- sampling Technique (SMOTE), time stretching, and time shifting, combined with RawNet-based waveform feature extraction integrated into the Audio Anti-Spoofing Using Integrated Spectro-Temporal Graph Attention Networks (AASIST) model. The method effectively captures subtle temporal and spectral artifacts, achieving an equal error rate (EER) of 10.39% on the SVDD SingFake 2024 dataset, outperforming baseline models. This work provides a robust solution to safeguard the authenticity of AI-generated singing voices in digital media.