The number of patients with dementia is increasing, and early detection of dementia is important to reduce the burden on medical institutions. Because Alzheimer's disease accounts for the largest proportion of causative diseases of dementia, this study focused on Alzheimer's disease and its preliminary stage, amnesic mild cognitive impairment. In this paper, we developed a machine learning model to detect nonactive for detecting cognitive impairment, focusing on motor evoked potentials before and after the application of transcranial magnetic stimulation to the brain. In this experiment, 24 experimental participants were measured for the motor evoked potentials. As a result of the experiment, the support vector machine was selected as the best weak learner, recording a sensitivity of 0.88 and a specificity of 0.63 in a measurement time of 15 minutes. Therefore, it is suggested that this model may be able to detect cognitive impairment.

Automatic music transcription (AMT) is a task that automatically generates music score from the sound of music. Pitch estimation is one of the most important technology in AMT, and its accuracy strongly affects the quality of AMT. Although its accuracy is improving thanks to deep neural networks nowadays, pitch estimation for multi-instrumental music is still a challenging task, because there are many difficulties caused by numerous and complicated timbre patterns in multi-instrumental music. In this paper, we propose a pitch estimation model based on the reconstruction of the spectrum of input sound which can reduce the effect of timbre complexity on pitch estimation. The experimental result indicated that the proposed model improved the average precision by 8.8 points compared to the conventional method.

In Japan, head-on collisions involving automobiles constitute approximately 30% of accidents between vehicles, placing them among the leading causes. Therefore, our research focuses on preventing head-on collisions by studying the detection of approaching vehicles using a Convolutional Neural Network (CNN) based on road environment sounds. To improve the detection accuracy of approaching vehicles using audio data, we believe it is desirable to handle longer-length input data. Therefore, we conducted verification of the impact on detection accuracy by varying the time length of input data to the conventional model. The results indicated that the conventional model may not effectively handle long-length data. Consequently, in this paper, we propose and evaluate a new CNN model that divides the input data at the central time point. As a result, the input data length of 2.49 seconds yielded the highest accuracy, which is 2.49 times longer than the conventional length. Additionally, the detection accuracy of approaching vehicles in the proposed model improved by about 5-10 percentage points compared to the accuracy of the conventional model.

Individual identification methods using ICT are required to reduce the workload of dairy farmers. Our research focuses on animal vocalizations as one of the resources useful for individual identification. In previous studies, the vocalizations during only hunger periods have been targeted. However, variations in acoustic information due to environmental and physiological differences may adversely affect individual identification. In this paper, we extract various kinds of acoustic information considering intra-individual differences of cattle and clear the acoustic differences in vocalizations between cattle states to identify individuals. As a result of the statistical tests, significant differences in acoustic information were observed between the hungry and estrus periods, revealing that cows exhibit acoustic variations depending on their states. We also conducted individual identification using i-vector and x-vector to account for such differences. The results showed that the x-vector achieved an accuracy of 92.2%. The x-vector demonstrated robustness, confirming its resilience to intra-individual differences.

Visually induced eye movements in larval fish, especially those in zebrafish have been intensively studied to understand neural mechanisms of sensory-motor transformation and motor learning. In this study, we developed a compact system for measuring eye movements in larval fish induced by visual stimuli using a small single-board computer, Raspberry Pi. The system features an integration of multiple displays for presenting visual stimuli and a camera for measuring binocular eye movements, both controlled by a Raspberry Pi.

Wireless local area networks (WLANs) have become increasingly popular and widely deployed. The medium access control (MAC) protocol is one of the important technologies in WLANs and directly affects communication efficiency. In this paper, we focus on the MAC protocol and propose a novel protocol in which network nodes dynamically optimize their backoff process to achieve high throughput in multi-hop adhoc networks. The distributed MAC protocol has the advantage of not requiring infrastructure such as an access point. However, under high traffic loads, the total throughput significantly decreases due to the hidden node problem, which requires improvement. Through theoretical analysis, we find that the average idle interval can represent the current network traffic load and can be used together with estimated number of neighbor nodes and hidden nodes for setting optimal contention window (CW). Through simulation comparisons with a conventional method, we show that our scheme can greatly enhance the throughput and the fairness in the saturated case.

A cyber physical system that enables to evaluate the electromagnetic (EM) field of the wireless communication environment in which mobile objects exist is proposed. The high-speed electromagnetic analysis is achieved by using the ray-trace calculation with huge and small numbers of polygons that electromagnetically represent stable and mobile objects like constructions and vehicles, individually. The variable rays generated by the equivalent sources on the mobile objects and the fixed rays interacting with the stable objects obtain the EM field that is useful for the actual wireless communication. According to fewer number of variable parameters, the proposed cyber physical system than that of the conventional one drastically reduces time to calculate the EM field for the mobile wireless communication.

Wire-free robot not using wires between links is proposed by utilizing wireless power transmission and communication. Attaching and detaching the links easily enable to install the arms into a facility on demand. Considering power supplied by the facility is limited, however, all of the robotic arms might not execute their task at same time. The worst case is whole system might shut down at the moment of task execution. This paper focuses on an operation planning method allowing the robot arms to execute their tasks under the total power constraints. Supposing that multiple arms control their end-effectors to follow given trajectories, instantaneous power can be discretized by considering only peak powers in each trajectory. Consequently, the operation planning problem is given as a solution to an integer programming by our method. As a merit of our method, the controllers programmed already on board of the arms need not to be changed, but to add a program solving the integer programming. The operation planning method is evaluated by a numerical simulation in this paper. This paper describes application requirements and its performance.

This study aims to develop a high-power acquisition system for nano-satellites. Microcomputers are utilized for power control of nano-satellites, considering size and power consumption. In this study, Perturbation & Observation (P & O), Particle Swarm Optimization (PSO), and Genetic Algorithm (GA) were implemented on Arduino UNO for Maximum Power Point Tracking (MPPT) control. The control processes were compared under conditions where the power characteristics exhibited both unimodal and multimodal functions. In MPPT control, key considerations include generated power, time to reach maximum power, and power fluctuations, all of which were focal points of our investigation. From experimental results, it was confirmed that PSO achieves maximum power even under conditions with a limited number of particles due to memory constraints. Additionally, it was observed that the operating point accurately tracked the maximum power for both unimodal and multimodal characteristics. The inference down from the experimental results suggested that PSO, where the Global Best is shared by all particles, proved to be an effective algorithm. These findings suggested the potential for achieving efficient MPPT control by using nano-satellites using PSO.

For the purpose of creating learning support tool for singing instruction, three acoustical parameters based on Singer's formant shown in classical singing voice is defined and discussed. Since Singer's formant is considered to be related to proficiency of classical singing voice, acoustical parameters that represent the relationship between them is required. SFR, Q, and the second convex point represent ratio, sharpness, and height of Singer's formant, respectively. SFR is defined based on the FFT spectrum. Q and the second convex point are defined based on the LPC spectrum. Three analytical questions as follow are examined and shown. (1) Definition of autocorrelation varies sharpness of spectral envelope, and it changes value of Q. (2) The second convex point is in the relation of the logarithm of Q. (3) To keep order of value of Q of various singings, adequate LPC analysis order is 12.

Human action recognition from video images generally requires a large amount of training data, but often only a limited amount of data is available. This paper proposes a data augmentation method using a generative model and a selector for skeleton-based human action recognition. Skeletons are extracted from video images and encoded into a pseudo-image with time-joint point coordinates. A generative model and a selector are trained on these encoded images for data augmentation. We conducted experiments using a diffusion model as the generative model and an Isolation Forest as the selector for a 21-class action classification problem on the JHMDB dataset. The results showed that adding the generated encoded images to the training data of the discriminator significantly increased the accuracy by 0.53 percentage points, from 55.05% to 55.58%, when the number of training data was 10 sequences per class. Furthermore, adding only the generated encoded images selected by the selector to the training data improved the accuracy by 0.90 percentage points to 55.95%.

This paper proposes discrete spider monkey optimization with dynamic multiple populations for a vending machine column optimization problem. For beverage vending machines, there is a need to reduce the number of vending machine restocking trips in order to tackle challenges of increasing fuel costs and decreasing the number of drivers. Therefore, it is necessary to extend the sell-out period of products in the vending machines to achieve the extension. It is possible to extend the period by selecting an appropriate product to each column in vending machines. The vending machine column optimization problem is a combinatorial optimization problem aimed at finding the optimal combination of columns for selling products. The effectiveness of the proposed method is confirmed through a comparison with conventional methods.

There is a lot of previous research on sales promotion strategy using digital signage. Maximization of expected sales is one of important topics on sales promotion strategy using digital signage. In the previous research, maximization of expected sales considering changes in purchasing intentions for one content selection has been studied. The previous method can not be applied to multiple content selections. In this research, maximization of expected sales considering changes in purchasing intentions for multiple content selections is studied. In the proposed method, the expected sales amount is maximized with respect to a Bayes criterion based on statistical decision theory. The effectiveness of the proposed method is shown by some computational examples. The expected sales amount of the proposed method is greater than that of a comparison target. In this research, the expected sales amount is maximized under the condition that all probabilities are known. But the probabilities are unknown in real cases. An expansion of this research with unknown probabilities is one of further works.

Risk assessment and management system for power lifeline against typhoon (RAMPT), developed by the CRIEPI, is a system capable of predicting the amount of damage caused to distribution equipment based on wind forecasts. This system is widely used in actual disaster situations. Generally, distribution equipment is affected by both direct forces caused by strong winds but and indirect impacts caused by fallen trees and landslides. Currently, RAMPT encounters challenges in predicting the damage caused by indirect impacts. Hence, this study develops a correction method for the prediction results of RAMPT by focusing on the indirect impacts causing damage to distribution equipment. An analysis of the relationship between the distribution equipment damage and indirect impacts using existing damage records revealed that the proximity between the equipment and trees and the landslide alert information from the Meteorological Agency presented a high correlation with the damage. Consequently, the application of correction factors based on information clarifying the relationship with damage improved the accuracy of damage predictions.

In this paper, we propose a method for classifying multiple gait patterns including abnormal movements when walking on stairs using micro-Doppler radar. We measured three types of movements while walking down the stairs, including an abnormal movement that mimics falling, and then predicted the classification using CNN based on the spectrogram derived from the STFT. As a result, we confirmed that the classification accuracies were more than 97%.

This paper introduces a parallel computation of Nested-Layer Particle Swarm Optimization (NLPSO) approach for investigating bifurcation points in continuous-time dynamical systems. The method offers efficient bifurcation point detection with limited prior knowledge and overcomes computational challenges posed by complex systems by employing parallel computing. The study provides a user-friendly tool for bifurcation point detection, eliminating the need for specialized knowledge or computing environments.