Showing 1–12 of 183 results

  • A Dual-Discriminator GAN for Sleep EEG Signal Synthesis

    Abstract: The interpretation of one’s overnight sleep process based on EEG signal is of importance for the inspection and treatment of various sleep-related disorders. The automatic sleep staging models have benefits as the assistant computerized tools to release the clinicians from the laborious task of manual sleep stage scoring. However, the issues of data insufficient and class imbalance are common for clinical data. The data problem is crucial to be solved when applying the automatic sleep staging models for real clinics. In this research, the network architecture of GAN (Generative Adversarial Network) is investigated by using one generator and two discriminators for the synthesis task of sleep EEG signals. The data augmentation performance by the dual-discriminator GAN and the sleep stage classification performance by combining with a typical machine learning classifier are evaluated on the sleep recording of subjects. The obtained results showed that the constructed dual-discriminator GAN is effective to generate samples which are closer to the time and frequency characteristics of real sleep EEG signal. It would be a useful method to solve the data problems for the training and optimization of automatic sleep stage classifiers in the field of sleep staging.
    Keywords: Generative adversarial network; Electroencephalograph; Data augmentation; Sleep staging; Dual-discriminator

  • Kendo Headgear Concussion Safety Evaluation

    Abstract: Concussions present a potential risk in various contact sports, including the martial
    art of kendo. While there are regulations implemented on the production quality of shinai,
    the bamboo-based swords used in kendo, there is a lack of active standards enforced by an
    independent governing body for kendo armor to prevent head injuries like concussions.
    Interestingly, two separate independent governing bodies exist to regulate and ensure the
    quality of commercially-sold shinai, but none for armor. This study aimed to answer two
    important questions: firstly, whether high-end kendo helmets offer superior protection
    against concussions compared to entry-level helmets, and secondly, whether adding extra
    padding inserts to the helmets enhances overall concussion protection. To conduct the study,
    we asked several kendo practitioners and one non-practitioner to deliver a series of head
    strikes using a shinai on a mannequin wearing different kendo helmets, both with and without
    additional protective padding. Our objective was to measure the force of each strike and
    assess the associated risk of concussion. The results indicated that both types of helmets
    sustained linear accelerations well below the threshold for the risk of concussion, which is
    set at 62.4g force. Notably, there were no statistical differences regarding impact forces
    received between the helmets (p-value equals 0.13). Interestingly, we found that commercial
    helmet padding inserts did not significantly reduce the risk of concussion. In fact, one of the
    commercially made helmet inserts performed worse than using a helmet alone (p-value is
    less than 0.01). In summary, our study suggests that investing in an entry-level kendo helmet
    with a 4-mm stitch pattern can offer comparable concussion protection to a high-end helmet
    with a 2-mm stitch pattern. As for commercially available padding inserts, their potential to
    provide additional concussion protection remains inconclusive and further studies are
    needed. Implementing a standardized evaluation process can aid consumers in making
    informed decisions when selecting protective gear to prevent concussions.

    Keywords: Kendo; Concussion; Sports Safety; Kendo Gears

  • The Development of a Smartphone Application based on Object Detection and Indoor Navigation to Assist Visually Impaired

    Abstract: A mobile application designed to assist visually impaired person (VIP) in performing daily activities such as grocery shopping. The system utilizes mobile application for object and location recognition, enabling users to locate and identify items in a supermarket easily. The application provides guidance through voice and vibration. We implemented a neural network and utilized Google’s GPS API in the application and conducted simulations in a supermarket environment to demonstrate its effectiveness. The proposed mobile application has the potential to significantly improve the independence and quality of life of visually impaired individuals.
    Keywords: Visual Impairment; Image Recognition; Indoor Navigation; Mobile Application

  • Development of a Japanese language learning support system for international students using video content

    Abstract: In recent years, globalization has progressed, and Japan’s international students have increased. However, many international students study while working part-time, and due to the impact of Covid-19, face-to-face conversation with people has become difficult. Therefore, regular study time alone has become insufficient for practicing the Japanese language. On the other hand, there are video distribution services that have become popular in recent years. Therefore, we thought we could create a language learning support system by using them. This research aims to develop a language learning support system that uses the subtitle function of a video distribution service to improve learning motivation and to solve the lack of time to learn a foreign language (the Japanese, in this case). This paper mainly reports on the development of the system by using those video content.

    Keywords: Language learning support system; Language Learning with Netflix; LLN extension; python VLC module.

  • EMG-Based Recognition Method of Finger Movement Impairment Level in Post-Stroke Patients Based on Fugl-Meyer Assessment

    Abstract:  The restoration of finger motor functions is considered difficult during rehabilitation due to the complexity of the underlying muscles. The Fugl-Meyer Assessment (FMA) method is used by doctors to manually assess the level of finger movement impairment. However, there is a risk of evaluation errors due to inherent subjectivity. Therefore, a new, more accurate method must be developed to predict the level of impairment. This study aimed to evaluate the impairment level of finger movement based on the FMA. EMG signals were recorded from four patients while performing seven movements, and feature extraction was performed. SVM and Random Forest were used to classify the level of impairment for each movement. The SVM model obtained good results in the fourth movement, with an accuracy of 91.7% and an F1 score of 0.78.

    Keywords: impairment level; finger movement; post-stroke patients; recognition.

  • Processing of Multi-valued Attributes Based on Sparse Matrix

    Abstract: Multi-value attribute has always been a difficult problem to deal with in machine learning. Most models are unique for data format matching. When it is multi-value, most models cannot be used directly. At the same time, a large number of multi-valued attributes will be encountered in the construction of medical model. These attributes often represent that patients have multiple symptoms. The processing methods of multi- valued attributes can be roughly divided into two categories, one is through data preprocessing, the other is through algorithm pattern matching. The solution to medical multi-valued attributes in this paper is mainly through preprocessing, from the perspective of multi-valued attribute representation and projection. The process is to use sparse matrix to represent multi- valued data, convert it to high-dimensional space, and then project it back to one dimension to complete the processing of such data.

    Keywords: Multi-value attributes; Sparse matrix; high- dimensional projection

  • Development of separation technology for dissimilar material bonding products of steel plate and CFRTP using the eddy current method

  • A Chinese text similarity algorithm based on Yake and neural network

    Abstract: Traditional text similarity algorithm has the disadvantage of  a large amount of text data and high complexity. Keywords are highly concentrated thematic ideas in the text. Extracting them can reduce the complexity of text similarity calculation. Therefore, this paper proposes a Chinese text similarity calculation method that integrates improved YAKE and neural network(YANN). With Aim to the  problem that Yet Another Keyword Extractor(YAKE) algorithm is not suitable for Chinese text keyword extraction, keyword candidate stage. First the new feature value of words is calculated by using word span, position, frequency, word context relevance, and the number of different sentences. Next we calculate the keyword score of each candidate word after synthesizing all the features values, and output the keywords in the ascending order of the score. Finally, the keyword set is inputted into the trained word2vec model for vectorization. Summation and averaging where the keyword vector values are derived from the trained word2vec model, and the similarity between different texts is calculated by cosine similarity. The experimental results show that the method proposed in this paper has better performance than other algorithms in Chinese text keyword extraction, and the similarity calculation results prove the merit of the method used.


    Keywords: Keyword Extraction; Word2vec; Text Similarity

  • Initial parameters of CNNs generated by Convolutional Sparse Representation with L1 error term

    Abstract: Convolutional Sparse Representation (CSR) approximates images with the convolutional sum of dictionary filters and corresponding sparse coefficients. To improve classification accuracy of Convolutional Neural Networks (CNNs), this paper proposes to use the dictionary filters generated by CSR as initial parameters of CNNs’ filters since the CSR filters express features of test images. Our method also estimates the error term of CSR with the L1 norm instead of the L2 norm to increase robustness against outliers in datasets for training. The results of experiments classifying CIFAR-10 show that the CNN using the initial parameters generated by the proposed method with the L1 error term shows the highest classification accuracy for small numbers of training images compared with the two methods: the proposed method with the L2 error term and the Xavier’s method.

    Keywords: Convolutional Sparse Representation, Convolutional Neural Network

  • Evaluation of the effects of cold and hot environmental temperatures on the distribution of whole facial skin temperature

    Abstract: During the Covid-19 pandemic, fever detection using infrared thermography became widespread. A person with a fever is detected based on the facial skin temperature measured in a non-invasive and free-of-restraint method. Recent studies have pointed out that the facial whole skin temperature, when measured immediately after entering a moderately moderate environment from a cold environment, is not practical for detecting persons with fever because it is greatly affected by the environmental temperature. On the other hand, the effect of cold and hot temperatures on the details of the entire face has not been evaluated. In this study, we compared the cold and hot environments and the acclimation to moderate temperatures to the effects of cold and hot environments on the whole face skin temperature distribution was evaluated in detail. The results showed that the periorbital area and side of the nose were least affected in the cold environment, and the side of the nose was least affected in the hot environment. And these parts are suggested to be suitable for core temperature estimation considering the environmental temperature.

    Keywords: facial skin temperature, environmental temperature, facial thermal image, core temperature

  • Development of flexible Text Input Device Based on Image Processing for Each Level of Disability Person

    Abstract: This research aims to develop a flexible input text device for physically disabled person using image processing. In proposed method we use a camera to detect the disabled person arm and hand movements via image processing technics, in order to identify these movements with exact disabled person intention on what he/she wants to convey. When the intention is detected, the disabled person can then input his/her intention or thought data through the device. The input method can be changed according to the user’s disability level and is expected to have a positive impact for the rehabilitation of the user.

    Keywords: Rehabilitation, Disability, Image processing, Input

  • Parallel Navigation of Multi-Drones using City Information for Search and Rescue Operations

    Abstract: The ongoing global warming causes an increased number of disasters around the globe and is going to have a severe impact on our society in the years to come. Thanks to the rapid development of technology, drones have emerged, and the Search and Rescue operation become effective and efficient. In this paper, we propose the assessment of multi-drones for Search and Rescue (SAR) operations. The proposed SAR uses an appropriate city model with a diverse population density where the multi-drones can be dispatched based on the city’s geographic information for the Search and rescue operation during a disaster.

    Keywords: Drone; Search and Rescue; Disaster; Operations