In this paper, we propose novel automatic piano music transcription methods which improve multi-pitch estimation accuracy. The conventional multi-pitch estimation method using low-rank non-negative matrix factorization (LR-NMF) cannot achieve a good accuracy because the method cannot detect true spectrum and analyze nonlinearity in sound. In the proposed method, the nonlinearity is analyzed using the convolutional neural network (CNN) or the convolutional denoising autoencoder (CDAE) as a post processing of LR-NMF. After the processing, we further improve the accuracy by the Hadamard product of the output from LR-NMF and that from CNN or CDAE. The performance of the proposed method is evaluated through computer simulation.

Optical flow estimation from an in-vehicle camera is an important task in automatic driving and advanced driver-assistance systems. However, there is a problem that optical flow estimation is mistakable with high contrast and high speed. Event camera can overcome these situations because it reports only the per-pixel intensity change with high dynamic range and low latency. However, the L1 smoothness regularization in the conventional optical flow estimation method is not suitable for radial optical flow in the driving scene. Therefore, we propose to use the focus of expansion (FOE) for regularization of optical flow estimation in event camera. The FOE is defined as the intersection of the translation vector of the camera and the image plane. The optical flow becomes radial from the FOE excluding the rotational component. Using the property, the optical flow can be regularized in the correct direction in the optimization process. We demonstrated that the optical flow was improved by introducing our regularization using the public dataset.

Higher education institutions or universities secure the quality of their education in order to build the confidence of stakeholders. In the mechanism of quality assurance, it is necessary to handle an enormous amount of text data related to their education activities. National Institution for Academic Degrees and Quality Enhancement of Higher Education conducts a research for quality assurance to analyze the text data of their admission, curriculum and diploma policies. An effective support system for text information processing by using information technology is desired. In this paper, we will consider the mechanism of preliminary system using character-level convolutional neural networks to judge the consistency between diploma policy and curriculum policy.

The quality of decision making by a RoboCup soccer agent depends on a path-planning and an evaluation function of a soccer field. In this paper, we employ a 5-layered neural network as the evaluation function. We examine the performance of the soccer agents with various features (ball coordinates and opponents' positions). We found that the neural network model helps soccer agents to imitate an expert team's decision making.

Pulsed electroacoustic (PEA) space charge measurements using a spin-coated Poly(vinylidene fluoride-trifluoroethylene) P(VDF-TrFE) thin film are demonstrated. The thickness of the P(VDF-TrFE) thin film is determined by a reflection interference method. The piezoelectric properties of the P(VDF-TrFE) film such as polarization hysteresis curve, residual polarization and coercive electric field are measured using a Sawyer-Tower circuit. The acoustic signals detected by 3-, 9-, and 30-µm piezoelectric films show strong thickness dependence, and the anti-resonance frequencies of the three films are evaluated by fast Fourier transform analysis. The space charge profiles on a 125-µm polyimide film surface are measured by PEA method. The space charge profile obtained by the 3-µm P(VDF-TrFE) film shows better spatial resolution compared with the results obtained other films.

This paper presents 2 improvements related to network construction exercise with network simulator IMUNES in 2018. Improvement 1 is that a method to copy strings printed in IMUNES's virtual terminals on FreeBSD OS to a window of application software on Windows OS is added to the exercise. The method make it easy for users to copy strings. Since IMUNES is an open source software, users can study network construction with the exercise and IMUNES with low cost. Improvement 2 is that how to edit an IMUNES network configuration file is added to the exercise. This method solves a problem that user can't save router configuration on IMUNES's virtual terminals.

Black lipid membranes suspending on a solid-state nanopore (nano black lipid membranes; nano-BLMs) is a functional application to detect single-channel activities with a long lifetime. However, folding and painting techniques, which are traditionally used to form nano-BLMs, only have a low success rate since its complicated procedure. Herein, we report on the self-spreading method on 50 nm pore size polycarbonate membrane filter to form nano-BLMs by a simple procedure. Our results show that the nano-BLMs have been successfully formed on the membrane filter. The spreading velocity of the nano-BLMs exhibits a square root behavior, v∼t^-1/2, which is in a great agreement with the previous research. The experiments with fluorescence recovery after photobleaching (FRAP) show the recovery of a photobleached area in our nano-BLMs. We also measure the elastic response of our nano-BLMs by using an atomic force microscope (AFM), and our results demonstrate that the lipid bilayer is spanning on the pore of membrane filter. We also discuss the effect of temperature and the roughness of supporting surface on the spreading speed of lipid bilayer.

In recent years, development of BCIs has been very active especially for patients with movement disorders such as ALS. However, current BCI’s require huge equipment, and sometimes users need training to achieve decent accuracy. In previous researches, AR-BCI (Augmented Reality BCI) has been developed to reduce such problems. Though AR-BCI can provide rich information comparing to previous systems, markers for AR and choices need to be prepared in advance. In this study, we developed an intuitive BCI system using Mixed-Reality (MR) and event related potential (ERP). MR-BCI recognize surrounding environment and provide appropriate options for users depending on their situation. In this time, we investigated how visual stimulus from MR system affects event-related potential, the influence of providing options of blind spots, and the influence of the style of virtual markers.

This paper describes control design considering magnetic flux characteristic uncertainty for magnetic levitation system with magnetic flux and current feedback. The position of the levitated object is calculated from the hall voltage and current of the electromagnet. We show the uncertainty for magnetic flux characteristic and design the stabilizing controller by proportional feedback with the hall voltage and current. Moreover, we tried to design servo controller by two-degree-of-freedom control by using a finite number of frequency response (FNFR) model including the uncertainty and confirmed the validity of the proposed method by experiment results.

ASPR based output feedback control with a predictive control as a feedforward input has been proposed for discrete time systems. The method can design a stable and simple output predictive control based adaptive controller with higher control performance for uncertain systems. In this paper, the method is expanded to continuous systems in order to deal with fast responding systems. In the proposed method, the controlled system will be augmented with a parallel feedforward compensator (PFC), witch is designed via a model free strategy, so as to render a minimum-phase augmented system with a relative degree of 1 for designing an output predictor and an adaptive output feedback control with a simple structure. The effectiveness of the proposed method using input/output data as a PFC design is confirmed through experiments via a magnetic levitation system.

We propose a method for reducing energy consumption of pumps in multi source district heating and cooling system. The method consists of an optimization technology and a control technology to compensate for modeling error. The proposed method automatically modify a parameters of the optimization model according to an actual valve aperture of a target system and prvent heat supply shortage or oversupply. The simulated results show the method is able to reduce 11.7% of the total energy consumption of pumps.

Nurse’s work is the task that leads to a serious accident with one mistake or miss, and the nurses are exposed to high stress. In particular, injection / pre-medication, tube connection, and falling are factors that lead to serious accidents and are considered to be a major load factor of nursing work. In order to reduce the burden of nursing work, we are working on the development of a sensing system to prevent fall accidents. Since fall accidents tend to occur when elderly people whose lower limb muscle strength has declined go to the toilet, we use the camera image to detect the end position, which is the initial posture of the patient’s landing movement. In this paper, we detected the sitting position of the patient by combining the detection result of the skeletal position of the patient and the detection result of the bed position. A simulation environment was constructed and the estimation accuracy of the end sitting position of the patient was evaluated by using the image taken of the scene where the patient and the nurse are active.

Refrigeration showcases are commonly utilized equipment in super markets and convenience stores to maintain the temperature and quality of products. Being also susceptible to fault events, the detection of symptoms of unusual operation is still difficult as only samples of normal behavior are usually available. This paper introduces a new use of autoencoders for this one class classification problem with only normal data. An unsupervised approach to cumulatively flag abnormal events is proposed based on ensembled autoencoders and compared with a deep learning counterpart, one-class support vector machine, and the multivariate statistical model standardly employed for fault analysis by the showcase industry manufacturer. Results showed the robustness of the method in flagging out-of-control samples, even when trained with raw sensor data without prior preprocessing.

Robustness and adaptability are necessary for metaheuristics because they are applied in various environment such as black-box optimization. In this paper, adding a parameter adjustment rule to Particle Swarm Optimization with rotational invariance using correlativity (CRI-PSO), we develop an adaptive CRI-PSO to improve the adaptability and maintain the robustness. First, using the swarm activity as an index evaluating the search state, we analyze a parameter of CRI-PSO based on intensification and diversification. Second, the parameter adjustment rule is based on evaluation and control of the search state. The rule controls the search state to realize intensification and diversification by adaptively adjusting the parameter. Also, the rule is designed so as to the adaptive CRI-PSO satisfies several transformation invariances of the solution space PSO not having. The performances (robustness and adaptability) of the adaptive CRI-PSO are verified through numerical experiments for typical benchmark functions comparing the adaptive CRI-PSO with three types of conventional PSOs.

A method to identify utility poles without being influenced by the measurement errors of Global Positioning System (GPS) is proposed for an Augmented Reality (AR) system for supporting visual inspection of power distribution facilities. When a system user views a pole through a camera mounted on a tablet, the pole is identified by the proposed method, and related information is overlaid on a live camera view. The proposed method narrows down potential target poles from plural candidates by checking positional relationship of the user's location and heading, the location (latitude, longitude) of each pole, and geographical information about buildings and roads. The user's location is estimated using not only GPS but also distance between the user and pole being inspected calculated by triangulation using the tablet's height and elevation angle. The proposed method is evaluated using a prototype at a residential area and an office area. The average of GPS measurement errors is 11.8m. The result of 90 times trials shows that the proposed method uniquely identifies the target pole at a success rate of 89.3%.

In this research, a compact feeding circuit for an antenna that achieves three-way switchable beam directions using lumped parameter elements was designed and constructed by replacing the transmission lines of the conventional circuit with a phase shifter using lumped elements. The validity of the prototype was evaluated through measurements. In addition, the area has been reduced to 1/216 the size of a conventional one.