[in Japanese]

View full abstract

In this paper, we discuss injurious bird recognition system that we have developed. Currently, crows (scientific name: Corvus), which are wild injurious bird, cause various kinds of damage in japan, such as damage to crops, as well as producing feces, noise, and garbage. Crows cause the largest amount of damage to crops among pest birds, and the amount of damage is enormous at around 40,000 tons per year. Our research group has been developing a repellent system that uses drones to drive away injurious bird, and we are now investigating the possibility of applying this system to crows. However, to apply this system to crows, it must be able to recognize the target crow with high accuracy, without background or other wild birds affecting its results. In this study, we developed a CNN-based crow recognition system and investigated its accuracy. As a result, we were able to obtain a 100% precision ratio for all classes in terms of the conformity experiment. In the recognition experiment, recognition rate is more than 87% for all classes. For the results, we used Grad-CAM to examine which parts of the crow were characterized, and we were able to confirm certain rules regarding the parts of the crow that were captured by CNN learning and the causes of misrecognition.

View full abstract

Hardware-based image processing is suitable for embedded systems because of its high speed and low power consumption. In addition, by using the technology of high-level synthesis, the creation of hardware can be done quickly and flexibly. In this study, we developed an embedded system with a camera that has a median-based dynamic background subtraction function. In this system, we introduced the concept of a ring buffer to run three pieces of hardware in parallel. As a result, when the image size is QVGA and the number of time-series images is 4, the maximum frame rate of the system is 194 fps. In addition, it was confirmed that the power efficiency of the system was much better than that of software processing.

View full abstract

In autonomous driving, longitudinal velocity control has attracted significant attention for research on autonomous driving. If the feedback gain of vehicle control can be tuned appropriately, the desired control response of autonomous vehicles can be desired. In this study, we proposed VRFT for vehicle controls. We extend standard VRFT an online type of controller parameter tuning method since vehicle dynamics are time-variant systems. This method is more practical than indirect methods which require a plant model identified by using the input and output data. because the proposed method can update PID gains online. Herein, the proposed method is verified through vehicle simulators. Compared to the fixed PID controller, the proposed method can realize desired velocity control performance when the actuator dynamics is varied.

View full abstract

Considering an aging society, it is necessary to make the electric wheelchair safer and easy to drive. The one of the important technologies to realize it is the tracking control. The tracking performance is affected by the environmental circumstances and the wheelchair dimensions alike. Therefore, the theoretical SIMULINK Model of wheelchair is developed, which makes it possible to develop efficiently the tracking control considering the effect of the wheelchair dimensions. And through the simulation using this SIMULINK Model, it is clarified that the gravity center of the chair must be considered to develop the algorithm.

View full abstract

We have developed a gait training rehabilitation robot “Akita Trainer” for hemiplegic patients. It has servomotors attached to the hip and knee joints of the exoskeleton, and driving it enables gait rehabilitation of the paralyzed people. As its core systems, we developed a variable assist system and an automatic assist adjustment system. The variable assist system based on compliance control can change the stiffness parameter of “Akita Trainer” to provide assist torque according to the degree of paralysis. The automatic assist adjustment system uses reinforcement learning to change the stiffness parameter to the optimal value for the paralyzed people while using this robot. We conducted an experiment with three healthy subjects to validate the combined system of both. We set up two walking patterns: , normal walking and pseudo-hemiplegic walking assuming right hemiplegia. The participants walked for 10 minutes on the treadmill at a speed of 0.8 km/h. The amount of assistance decreased for normal walking and increased for pseudo-hemiplegic walking, indicating that the optimal amount of assistance can be provided according to the walking patterns. By using this system, it is possible to perform gait rehabilitation with the optimal assist torque for the hemiplegic people.

View full abstract

Recently, wind power generators are being introduced all over the world. However, in Japan, the damage caused by lightning strikes wind power generators is not small, and their introduction has been stagnant. In this study, the surface discharge phenomena were investigated to reduce the damage caused by lightning strikes on wind power generators. Experimental results of surface discharge on the effects of surface conditions on wind power generator blades have been reported.

View full abstract

The practical application of image processing for visual inspection is being promoted, but there are some challenges that cannot be detected only by rule-based methods. In this paper, we evaluate rule-based and machine learning methods for visual inspection of casting parts. Rule-based is a method in which an operator interprets the features of an image and sets rules for judgment such as thresholds. Machine learning, on the other hand, is a method that automatically interprets features from a large amount of data and determines decision rules without using explicit instructions. As a result of the comparison, it was found that machine learning is suitable for the visual inspection of cast parts.

View full abstract

Currently, the estimated value of the effective reproduction number (ERN), which is an index for grasping the COVID-19 infection status, is used for important planning and evaluation of infection prevention measures. Since ERN in the Sequential SIR model fluctuates in multiple dimensions due to changes in the surrounding environment, it is difficult to set the appropriate accuracy of the uncertainty region of the estimated data. The challenge in this study is to build a mathematical model of infectious disease according to the characteristics and data characteristics of the infectious disease and select an appropriate estimation method. Highly accurate quantitative research that analyzes the validity of "how infectious diseases prevail" from an academic point of view is the key to prediction and estimation in appropriate infection situation analysis. In this paper, we will apply the "multivariate analysis method (T Method)" that can evaluate important factors related to ERN prediction and estimation and efficiently process multidimensional information, and explore the construction of a new sequential model of infectious diseases. By applying the multivariate analysis method, we will clarify the important factors related to the prediction and estimation of the infection situation and their contributions, and approach the creation of more permanent infection prevention measures.

View full abstract

There have been many accidents around wind turbines in Japan, caused by lightning strikes due to weather, climate, and location, etc. The regular inspection method on wind turbines after a lightning strike is generally to stop the operation of the wind turbines for a certain period of time for visual checks. This situation requires another method that will allow the easy detection of lightning strikes. We installed an acoustic emission sensor on the enclosure of a large wind turbines and observed a shock wave produced at the moment the lightning struck the large wind turbines. We successfully detected the lightning strikes and are reporting the details in this paper.

View full abstract