We aim at engineering applications of reservoir computing using hardware chaotic neural networks, including associative memory recall. The reservoir layer used in reservoir computing is networked and constructed using Pulse-type Hardware Chaos Neuron Models (P-HCNM). The structure of the reservoir layer is simple, which is advantageous for hardware implementation. By inducing chaos in the reservoir layer, it is possible to use the "chaotic edge" where the reservoir reaches its highest efficiency. It has also been reported that incorporating self-correction within the reservoir layer increases the efficiency of the task.

In this paper, we constructed a hardware small-world neural network using a synaptic model with spike timing-dependent synaptic plasticity (STDP) and a Gap Junction model. As a result, it is clarified that all cell body models with synaptic model connections show chaotic firing by simulation at the same time, and that the STDP model enables learning while keeping the chaotic phenomena. In addition, comparison with the firing of cell body models coupled only with synaptic models suggested that the Gap Junction model works significantly in inducing chaos in neural networks.

The sensation of drowsiness experienced around 2 PM, referred to as post-lunch dip (PLD), induces a decline in brain functions including attentiveness and levels of alertness. Such transient brain function declines instigated by PLD have been identified as potential catalysts for human errors. Therefore, it is important to demonstrate optimal intervention methods and preventive techniques to remedy this phenomenon. Our research experiment examined the influence of variations in mastication frequency during white rice consumption on blood glucose levels and event-related potential (ERP). Our study was conducted on fifteen young male participants. The measurements comprised of P300, CNV, blood glucose levels, and a questionnaire on sleepiness. We recorded ERP pre-consumption, immediately post-consumption, and 40 and 80 minutes after the meal. We employed packed rice as the dietary load and controlled the mastication frequency per mouthful at either 10 or 40 chews. Our results indicated that at 80 minutes post-consumption, when the mastication frequency was 10 chews compared to 40, there was an extended P300 latency, reduction in late CNV, increased reaction time, and decreased accuracy rate (p<0.05). These findings suggest that intensifying masticatory activity could effectively mitigate the transient decline in brain functions caused by PLD.

Drivers show two types of stress-coping responses: active coping responses to stressful stimuli that they actively cope with and passive coping responses to stressful stimuli that they have no choice but to endure. Considering that the two types of stress-coping styles of drivers can be used as indicators of safety and comfort evaluation at all SAE levels, a method must be established to distinguish between them. Stress-coping styles can be distinguished based on hemodynamic variability; however, existing hemodynamic measurement methods require contact and time. In this study, we measured facial skin blood flow using a noncontact method by capturing near-infrared facial images using near-infrared light, which has a high depth of penetration into biological tissues. Furthermore, we applied sparse modeling, an information extraction technique, to near-infrared facial images to extract skin blood flow features related to stress-coping styles and attempted to distinguish stress-coping styles as quickly as possible using a single image. Consequently, we identified skin blood flow characteristics of the nasal apex and cheeks related to stress-coping styles and developed a noncontact, rapid discrimination method with an accuracy of 83.1%。

We developed an assist robot that changes its assist force in real time according to the lumbar load estimated from the load information on the hand measured using a hand sensor device and the posture information. Furthermore, using the developed assist robot, the effect of the load-following control on muscle fatigue was verified. The load was set at 6 kgf and 1 kgf, and the amount of muscle activity in the lumbar region was measured using a muscle potential sensor during continuous flexion-extension exercises. The central frequency of the power spectrum was calculated as muscle fatigue, and its time trend was obtained. For comparison, similar experiments were also conducted without an assistive robot and with an existing assistive robot. Consequently, when load-following control was used, muscle fatigue was reduced compared with existing assist robots in which the assist force was excessive in relation to the load. This shows the importance of load-following control that is necessary to control the assist force according to the load when the assist robot is worn during work in which the load changes in a complex manner.

Sleep apnea syndrome (SAS) is a disease that causes apnea during sleep, and since it correlates with lifestylerelated diseases, early detection and early treatment are important. Polysomnography (PSG) is generally used for the definitive diagnosis of SAS, however, since various devices are attached to the patient, the effect of observer effect is inevitable. In addition, since the disease occurs during sleep, it is difficult to be aware of it, and as a result, it is estimated that 15% of potential patients do not undergo PSG. This paper proposes a simple, noncontact and non-constraint respiration estimation method using Depth data, which is distance information obtained by a Depth camera and recorded in a two-dimensional plane. Frequency analysis of the Depth values is used to detect the subject's position, breathing frequency, and body motion, and it is confirmed that the method can estimate respiration while continuously tracking the subject, even if the sleeping posture changes.

In recent years, with the widespread adoption of remote work alongside traditional office work, the challenge of sharing the office atmosphere remotely has become evident. This difficulty has raised concerns about a lack of solidarity, a sense of belonging, and an increase in psychological stress among remote workers. People can imagine the atmosphere of a place from its sounds, but it is not clear from which sounds they imagine the atmosphere. In this study, we attempt to analyze the sound field context information of office spaces in order to clarify the sounds that constitute the atmosphere. First, we categorize the sound field context based on voice recordings made in a simulated office environment. Next, we analyze the influence of the sound context information and individual characteristics on the impression based on the sound impression evaluation experiment. Finally, we conduct a psychological experiment to evaluate the categorized sound field context information. The results revealed that the sound field context information can be categorized by individuality and context, and that sounds with higher resolution in these two areas are more likely to leave a lasting impression on people. On the other hand, we found that sounds that were too explicit were perceived as bothersome by people at work and decreased their subjective work efficiency compared to quieter conditions.

There is a high demand for a safe and cost-effective method to inspect the tightness of bolts remotely, which are widely used in power plants. Since it is known that rigidity decreases of a structure due to bolt loosening can be reflected in natural frequency of the structure, many methods for detecting bolt loosening through natural frequency have been proposed. However, most of those require the installation of contact sensors or markers, making them difficult to use when the main vibration point is at a high location. To overcome this problem, we utilized a video-based vibrometry technology which enables to measure vibration from a distance without any pre-installed sensors or markers and measured vibration of a substation’s steel structure under impact excitation. The result suggested that the technology could remotely detect the decrease in dominant frequencies of the neighboring parts of intentionally loosened bolts, as well as the disappearance of the peaks.

As WHO advocates well-being, people must be physically, mentally, and socially satisfied. Japan is aiming to realize well-being under a policy called Society 5.0. In this background, this study proposes a basic design scheme of a personal-fit control system in a feedback loop composed of a machine and a human. The personal-fit control system automatically searches for the optimal characteristics of a system to maximize a user's personal-fit index based on an extremum seeking control approach. The personal-fit control system automatically can achieve a vehicle behavior that a user desires. The simulation results show that a vehicle model can automatically adjust its system parameters to match the internal model that a driver assumes.

The study considers a direct controller parameter design for regulatory control using one-shot experimental data contaminated with random disturbances. The proposed approach employs an idea of reference model matching, and provides an estimation method for simultaneous disturbance and the desired controller that makes the closed-loop system match with the reference model using a predictive error method (PEM). The study also presents an identifiability condition that assures local uniqueness of the estimated parameters. Finally, a numerical example demonstrates effectiveness of the proposed approach.

This study considers data-driven controller parameter tuning for nonlinear systems using one-shot input-output data. The study addresses the model reference problem for a nonlinear system described by the parametric strict-feedback form and determines the controller parameters from one-shot experimental data. In the proposed method, the idea of the VRFT (Virtual Reference Feedback Tuning) is applied to the tuning of nonlinear controller parameters. Finally, this paper shows numerical simulation results to demonstrate the effectiveness of the proposed method.

This paper proposes a bilateral control with a tension control considering joint torque for a tendon-driven robotic hand. This robotic hand has two joints and two degrees-of-freedom, which is driven by three tendons. Tendon arrangement in the robotic hand is designed in terms of joint torque output. In tendon-driven mechanism, tension control is needed for power transmission. This research achieves a tension control considering the joint torque that fluctuates depending on the hand posture. In addition, the bilateral control with tension control is implemented. The effectiveness of the proposed method was further demonstrated through experimental results.

In geological surveys, a borehole camera is used to photograph the vertical cylindrical borehole-wall to investigate underground cracks. Currently, the identification of cracks from borehole-wall images is performed visually by skilled workers, which requires a great deal of time and effort. In this study, we use deep learning to detect sine-curve-like cracks from borehole-wall panoramic images. We designed a two-class classification model that discriminates the presence or absence of cracks using existing network architectures. Furthermore, we introduced a new data augmentation technique called “CyclicShift”, which takes advantage of the unique properties of borehole-wall panoramic images. Through experiments using our own dataset, we showed that both WideResNet and ViT achieve over 98% accuracy under the limited condition of a single crack in one image. Additionally, we confirmed the effectiveness of data augmentation and fine-tuning of pre-trained models. We also demonstrated the potential of using Grad-CAM to locate the positions of cracks.

In Japan, despite seat belts being mandatory for drivers, some choose not to comply. This non-compliance is typically identified through visual inspections by police officers. However, Japan's population decline has facilitated a growing need for automating this process using camera and vision technologies. This study explores the optimal combination of semantic segmentation models and loss functions for seat belt detection in images. We created a dataset using various car models in outdoor settings and evaluated the performance of all combinations of nine models and five loss functions. Our findings indicate that the UNet++ model paired with the Lovász loss function delivers superior performance.

Hypertension is a risk factor for cardiac and cerebrovascular disorders, and routine blood pressure monitoring is important for its early detection. The previous study that attempted to detect hypertension by applying CNN to facial visible images found that the problem was that features other than physiological responses, such as facial expressions, were mixed. Thus, we applied sparse coding to the facial visible images. However, we were able to extract features related to acute blood pressure fluctuations, we were unable to obtain sufficient accuracy. One of the reasons for the low accuracy is that the visible band is a wavelength band that is easily affected by ambient light. In contrast, the near-infrared (NIR) band is highly permeable to biological tissues and reduces the influence of external light. In this study, we attempted to detect hypertension by applying sparse coding to facial NIR images, which can capture blood flow fluctuations deep inside the body, in addition to facial visible images. By using different wavelength bands, information from the surface to the depth of the living body can be obtained, which is expected to improve the accuracy of hypertension detection. Besides, the dimensionality reduction methods, principal component analysis (PCA) and t-distributed stochastic neighbor embedding (t-SNE), were used to compare with sparse coding. As a result, a hypertension detection accuracy of 81.0% was obtained when visible images and images obtained from a Si NIR camera sensitive to 760 to 900 nm were used together. This result suggested that the detection accuracy can be improved by using multiple wavelength bands together.

Nicotiana Benthamiana, a candidate plant for plant-derived vaccines, is considered suitable for cultivation in a multi-stage plant factory. To increase area efficiency, we investigated the possibility of growing Nicotiana Benthamiana at a lower plant height. The results suggested that low plant height could be achieved by electric irradiation.

Cattle blood tests are used for feed adjustment and early detection of disease. Conventional cattle blood tests pose infection risks and handling challenges. This paper proposes estimating blood glucose levels in cattle via near-infrared spectroscopy (NIRS) and a CNN model. Showing an accuracy of 94.67%, it demonstrates its effectiveness as one of the metrics in veterinary medicine.

We propose an eye-gaze input system that utilizes a laptop PC and its inner camera. This system can discriminate the user’s eye-gaze direction by using Convolutional Neural Network (CNN) or Vision Transformer (ViT). In this paper, we present the results of a comparison of the newly created eye-gaze direction discrimination model of ViT and the past model created by a CNN. We evaluated the accuracy of discrimination models created by ViT and CNN through the experiments. As a result, the ViT model has higher accuracy than the CNN model in discriminating the center direction.

This paper presents we added a feedforward circuit to improve the signal quality by adding a feedforward circuit to the third-order distortion caused by increasing the modulation index of the LiNbO₃(LN) optical modulator, and described its effect. Furthermore, we applied this optical modulator distortion reduction experimental device using this feedforward to a 5-channel simultaneous BS television information signal modulation system, and described its characteristics.

In this study, we propose an optical method for noninvasive and quantitative measurement of blood viscosity in extracorporeal circulatory device. Using self-made simulated blood images, we investigated a method to detect aggregation regions by extracting feature points. Then, we investigated the estimation of aggregation number and aggregation diameter using clustering. As a result, it is possible to detect aggregations and discriminate the number of aggregations and their size.

Deep Brain Stimulation (DBS) plays an important role as a promising clinical treatment for drug-resistant neurological and psychiatric disorders such as Parkinson's disease (PD), epilepsy, and major depression. Recently, closed-loop DBS using classical control theory such as proportional-integral (PI) control has been proposed as an on-demand stimulation strategy to reduce the power consumption of the stimulator. However, while DBS with variable pulse amplitude and constant pulse frequency has been studied, the performance of DBS with variable pulse frequency and constant pulse amplitude has not yet been clarified, leaving room for further investigation. In this study, we investigate how DBS with variable pulse frequency by PI control can suppress abnormal β oscillation derived from PD by computer simulation using a biophysical model of the cortico-basal ganglia-thalamus neural network model. Along with that, the optimal range of PI control parameter combinations was estimated. As a result, the DBS with an appropriate combination of PI control parameters suppresses the pathological β oscillation and reduces the power consumption. These results indicate that the proposed stimulating strategy with a variable pulse frequency can be applied to controlling PD induced beta power and can be more efficient than open-loop DBS using current technologies.