Program

2020 2020 IEEE International Conference on Consumer Electronics - Taiwan (ICCE-Taiwan)  
This paper proposes a probability density function model based on deep learning to analyze the relationships between the number of call arrivals and vehicle speed. Furthermore, a vehicle speed estimation method based on deep learning is proposed to estimate vehicle speed in accordance with the number of call arrivals. A traffic flow estimation method is proposed to estimate traffic flow in accordance with the number of normal location updates. Finally, a traffic density estimation method is
more » ... osed to estimate the traffic density in accordance with the estimated vehicle speed and the estimated traffic flow. In experiments, the simulation results showed that the accuracies of estimated vehicle speed and estimated traffic density are 96.36% and 96.45%, respectively. In this paper, a data filtering method is proposed to filter out the interference of mobile network signals and location errors from global positioning system in training data for the improvement of mobile positioning. Furthermore, a mobile positioning based on deep learning is proposed to analyze the received signal strength indications (RSSIs) from cellular networks and WiFi networks for determining the location of mobile station. In practical experimental environments, a case study of Fujian University of Technology is selected to evaluate the proposed methods. This study collected 3655 records which include 77 different base stations, and 1947 different Wi-Fi access points. The experimental results showed that the average location errors of the proposed method was 4.32m. The latest research from Rockefeller University in the United States in 2014 found that humans can smell one trillion fragrances, and smell is one of the best ways to wake up memory. HERMES's exclusive perfumer, Elena, mentioned in his book "Perfume Master's Diary" that in life, smell is everywhere. Fruit vendors in the market, food from exotic restaurants, and so on, may all evoke the waves of imagination and become inspiration for making fragrance. As for the method of remembering scent, Elena used drawing to visually assist olfactory association in the diary to help herself remember the scent. In this research, we applied the GPT-2 algorithm in natural language learning in deep learning to allow computers to learn the method of perfumery by crawling the Internet. Let the computer learn how to scent and produce a formula. Although the computer does not have a sense of smell, it can still summarize and infer through various documents to produce the fragrance in memory for subsequent related research. It is a challenge to predict the long-term future data from time series data. This paper proposes to use a Transformer with soft dynamic time wrapping for early stopping criteria, called a soft-DTW Transformer. Our experiment in an open-source dataset HouseTwenty shows that the average prediction error rate with soft-DTW Transformer is 27.79%, greatly reduced from 45.70% for using SVR, a common time series method. The video streaming platforms have recently evolved to provide diversified services, including live video, video on demand, and TV shows premiere. The traditional unicast method may cause a waste of bandwidth when many users watch the same video contents at the same time. In this study, we realized a hybrid unicast and multicast adaptive streaming system. The system uses ROUTE (Real-time Object delivery over Unidirectional Transport) combined with the MPEG-DASH (Dynamic Adaptive Streaming over HTTP) to simultaneously satisfy the needs of unicast and multicast users. We use the SHVC (Scalable High-efficiency Video Coding) to generate high-quality video of different bitrates. The QoE (Quality of Experience) function is measured by bitrate, PSNR, and server bandwidth. The proposed adaptive algorithm dynamically allocates the bitrates of unicast and multicast users and achieves optimal average viewing quality. Experimental results show that the proposed algorithm effectively improves the system's QoE under various bandwidth models compared to the conventional method. In the hybrid multicast and unicast scenario, the PSNR for the stable, incremental, and decreasing bandwidth models are increased by 0.79 dB, 4.92 dB, and 7.88 dB, respectively. The average quality switching frequency of users is also decreased. Distributed processing technology in wireless sensor network (WSN) has attracted attention because of the performance improvement of sensor nodes. Although the conventional methods divide and allocate computational processing of machine learning to sensor nodes, appropriate allocation has not been realized from the viewpoint of the whole network. In this paper, assuming that multiple machine learning processes occur simultaneously, we propose a processing division and allocation method to equalize processing load on sensor nodes. The evaluation results show that the method can almost fairly distribute the load to sensor nodes. 9:30 Multipath Routing Method for Large File Transfer with Time Constraint Rintaro Sozu and Kazuhiko Kinoshita (Tokushima University, Japan) In recent years, time-constrained file transfer model receives much attention. It promises to complete the requested transfer by its deadline or rejects. On the other hand, multipath routing using two or more routes at the same time is effective for load balancing and throughput improvement. In this paper, we propose to apply multi-path routing technique to the file transfer with time constraint. Finally, the effectiveness of the proposed method was confirmed by simulation experiments. In the fifth generation (5G) mobile network, it is necessary to realize ultra-low latency communication. Currently, multi-access edge computing (MEC) is attracted attention as a technology for realizing the ultra-low latency communication by using MEC servers that are placed near users. In this paper, we implement a dynamic task assignment for multi-user information sharing in smartphone application that use MEC and cloud servers. In our implementation, the MEC servers dynamically decide whether each task is processed in the MEC servers or the cloud servers. We evaluate the performance of our dynamic task assignment in our experimental system. From experimental results, we show that our task assignment can decrease a response time of each task according to the processing capability of MEC servers. Network function virtualization (NFV) provides several types of network functions by implementing virtual network function (VNF) on a server. In a service chain that utilizes some VNFs, the number of VNF instances and route selection have to be managed for the service chain dynamically according to the amount of traffic. In this paper, we propose an optimal VNF placement and route selection with model predictive control (MPC) for multiple service chains. The proposed method utilizes the predicted amount of injected traffic for each service chain to decide the optimal number of VNF instances. In addition, this method decides the optimal route for each service chain. These processes are performed by using MPC appropriately. We evaluate the performance of the proposed method with simulation and investigate the effectiveness of the proposed method. The piezoelectric element resonance is used to generate energy for cardiac catheter ablation systems. The design of piezoelectric vibrators is closely related to the two major directions of material selection and geometry. In terms of standard AC to DC output power, its size is closely related to the external source form, the electromechanical properties of the vibrator (mass, damping, stiffness, forceelectricity coupling, parasitic capacitance) and the external load connected to it. Bio-living pulsed light-driven near-infrared image recognition system can assist medical personnel to quickly and accurately find and locate veins during venipuncture, especially for patients with difficulty in locating blood vessels such as young children, obesity, edema, and hairy, which can help in order to improve the success rate of venipuncture, reduce patient suffering, and improve the doctor-patient relationship. The heart rate sensing module LED estimates relatively accurate heartbeat data for hundreds of flashes per second. In most cases, PPG technology uses infrared for sensing. When the data obtained by infrared sensing fails to meet the requirements, it automatically switches to acquiring heart rhythm data using the PPG scheme by increasing the pulse driving frequency. This article explores the transmit energy of a linear amplifier used in medical Doppler driver systems. In order to verify the mechanical pulse strength of the ultrasonic wave generated by the ultrasonic probe excited by the linear amplifier, the actual weight of the transmitted wave has been measured using an electronic balance, and the relationship between the weight and the amplitude of the transmitted signal. In addition, the acoustic pressure fields was measured using a calibrated hydrophone. The intensities of energy density were calculated and the quality of underwater signals has been verified. 10:00 Quick Placement and Precise Positioning DNA Liquid Jet Technology Jian-Chiun Liou and Zhen-Xi Chen (Taipei Medical University, Taiwan) The main working methods of liquid inkjet printers are divided into thermal foam type and ceramic piezoelectric type. The technology of warming liquid jet to maintain stable viscosity and surface tension has also been applied to DNA liquid jet. This study is described Quick Placement and Precise Positioning DNA Liquid Jet Technology. Observed the spray DNA under the microscope and the clear image of the DNA droplets. Gait phase detection with convolution neural network provides accurate classification but demands high computational cost, which inhibits real time low power on-sensor processing. This paper presents a segmentation based gait phase detection with a width and depth downscaled U-Net like model that only needs 0.5KB model size and 67K operations per second with 95.9% accuracy to be easily fitted into resource limited on sensor microcontroller. A4: Advanced Computing Systems and Applications I Room: Lyon D Chair: Masaru Fukushi (Yamaguchi University, Japan) 9:00 Interleaved Non-Binary LDPC Code for Synchronization Error Correction in DNA Storage Haruhiko Kaneko (Tokyo Institute of Technology, Japan) DNA storage devices are generally prone to synchronization errors, and hence these devices will require strong synchronization error correction coding to improve the reliability. This paper presents a insertion/deletion/substitution (IDS) error correction coding based on interleaving of non-binary LDPC codes. We employ an iterative decoding using a belief propagation on drift factor graph and a non-binary sum-product algorithm. This decoding successively recovers synchronization from both ends to center in a received word. Simulation results show that the presented coding of block size \{4096 \times 100\} with rate-\{3/4\} gives decoded block error rate \{2.0 \times 10^{-3}\} for IDS channel of insertion/deletion error probability \{2.1 \times 10^{-2}\} and substitution error probability \{1.0 \times 10^{-2}\}. This paper simulates the influence of recognition performance of using a phase-change memory as weight of a recurrent neural network. In the first experiment, a neural network did not learn well due to nan error caused by write error in phase-change memory. In the second and later experiments, there was no correlation between the write error rate and the validation loss. However, when the learning results were output, higher write error rate caused the less successful the learning. 9:30 Goal Recognition Using Deep Learning in a Planetary Exploration Rover Developed for a Contest Miho Akiyama and Takuya Saito (Shonan Institute of Technology, Japan) We participated in the "A Rocket Launch for International Student Satellites (ARLISS)" competition in which original design planetary exploration rovers competed to reach close to the target using autonomous control. In this competition, the rovers of various teams approached the target position using the global positioning system (GPS). However, they could only approach to within a few meters of the target due to the GPS positioning error. Our rover recognized the red traffic cone, placed at the goal point, by its color and in the Tanegashima Rocket Contest 2018, the rover was controlled to the point where the distance to the goal was 0 m. However, image recognition of goal objects by their colors suffers from the problem of unstable recognition due to changes in ambient lighting, which occurs due to, for example, weather changes. We therefore attempted to resolve this problem by employing deep learning. However, a considerable amount of calculation time is taken by a general deep learning model to run on a small planetary exploration rover computer and thus cannot be applied as it is. Therefore, we proposed a deep learning model with a short calculation time and high recognition accuracy. Using the proposed method, a recognition rate of over 99 % could be achieved in a few seconds. Furthermore, we won the contest by demonstrating the effectiveness of the rover using the proposed method and thus proved the effectiveness of this method. 9:45 Influence of Radio Waves Generated by XBee Module on GPS Positioning Performance Miho Akiyama and Takuya Saito (Shonan Institute of Technology, Japan) In the past, we had positioning problems on a rover equipped with an XBee module. Therefore, we investigated what caused these GPS positioning errors. As a result, we found that the XBee module of our rover affects GPS performance. Furthermore, we found that the precision of the floating-point calculation is not enough for Arduino Uno. We investigate the frame subtraction method in the research of the security surveillance system for old people by artificial intelligence using deep neural networks. The results in low-illumination images become smaller when the conventional inter-frame differencing method is applied. Therefore, we propose an optimal method of inter-frame differencing in various illuminance environments. The accuracy of inter-frame differencing improves when the proposed method using contour processing is applied. In this paper, we propose a method for automatic labeling of captured video images by using parametric eigenspace method for AI learning data. The experimental results showed that about 70% of the categorized data was labeled correctly. We also tried to improve the accuracy of labeling by extracting the moving object regions. The experimental results showed slight improvement in the accuracy of labeling. In this paper, we study a real-time traffic condition assessment method on the basis of taxi GPS data of Dongguan city by a new sensing form-crowdsensing. First, data preprocessing is used to clean invalid and noisy data. Second, we propose a road network extraction solution by mixing grid method and least square curve fitting. Third, we present a geometric matching based map matching solution. Finally, a velocity parameter estimation scheme based on self-adaptive traffic flow velocity is employed to assess traffic condition. A series of simulation experiments are carried out to verify the effectiveness of our proposed solutions. To better protect users' privacy, various authentication mechanisms have been applied on smartphones. Android pattern lock has been widely used because it is easy to memory, however, simple ones are more vulnerable to attacks such as shoulder surfing attack. In this paper, we propose a security rating evaluation scheme based pattern lock. In particular, an entropy function of a pattern lock can be calculated, which is decided by five kinds of attributes: size, length, angle, overlap and intersection for quantitative evaluation of pattern lock. And thus, the security rating thresholds will be determined according the distribution of entropy values. Finally, we design and develop an APP based on Android Studio, which is used to verify the effectiveness of our proposed security rating evaluation scheme. In this paper, we propose a fine-grained PM2.5 detection method based on crowdsensing technology. Firstly, we perform dark channel processing on sky images which have been collected by mobile users through APP. Secondly, we adopt a train model based on Tensorflow and Keras architecture, and utilize neural network to implement PM2.5 feature extraction and detection. Finally, a series of experiments have been carried out based on a large-scale real dataset to verify the performance of our proposed detection method. Machine learning technology has been widely used in the field of auxiliary diagnosis. However, existing models ignore the association between words. Therefore, performance improvement is limited when the scale of data is expended. To solve this problem, we use the cw2vec model to process the Chinese medical record text, combine it with the fusion model of Bidirectional Long Short-Term Memory and conditional random field, and propose a new model. This model makes full use of the word embedding technology and improves the existing neural network by combining the semantic connotation of Chinese. The experimental results show that the improved model has higher disease identification rate than the existing models under the same data scale. 10:00 Research on NDN -based Vehicle Network Cache Strategy Chunling Chen, Haoyang Xi and Yongan Guo (Nanjing University of Posts and Telecommunications, China) With the rapid development of Internet of Vehicles (IoV), TCP/IP-based IoV encounters problems in mobility, flexibility, and network latency. Thus, this article introduces NDN-based IoV and the related cache strategies, analyzes the challenges in developing NDN-based IoV, and compares several representative cache strategies. In this paper, the designs of graph filters using least-squares (LS) method with parameter norm penalty are studied. First, the design problem of the graph filter which can be used to process irregular sensor network data is described. Then, the conventional LS method is employed to design graph filter and its numerical stability problem in high-order filter case is presented. Next, to solve numerical stability problem, the LS method with parameter norm penalty is applied to design the graph filter. Finally, the design examples are demonstrated to show the effectiveness of the proposed design method. In this paper, the graph construction of signal de-noising method using Laplacian matrix is investigated. First, the four-step procedure to obtain a graph signal representation is described. Then, the details of graph signal de-noising method using Laplacain matrix are presented. Next, four kinds of graph constructions with different connectivities are studied including Hamiltonian cycle, spanning tree, Delaunay triangle and complete graph. Finally, the real temperature data is used to evaluate the performance of the graph signal denoising method and to show which graph construction is a better choice according to the value of signal to noise ratio (SNR). This study presents a preliminary study of investigating the interface design of smartwatch application menu and highlights issues at the hierarchical level of a classified application menu interface. After investigating the current smartwatch application menu interface and interviewing interface design experts, the results show that current application menu interface are designed variously according to different product positioning. Therefore, there is no universal and proper information architecture are applied to take care of numerous application menu interface arrangement. Future work will focus on investigating the possibility of designing interface by classifying the application menu to improve the user experience. With the rise of mobile phones and the popularity of the Internet, 5G, and e-commerce, there should be more various tasks processed on smartphone, which requires a good scheduling strategy to handle running processes. Compared with a single core, one of the biggest advantages of multi-core is the efficiency of running processes. In the multi-core, each core is responsible for serving a process at each moment, hence the performance of smartphone executing programs can become much better. This paper focuses on designing a branch-and-bound strategy for process scheduling in multi-core smartphone. Since the shifts of cores must be allocated by the operating system (like Android or iOS), the scheduler in the operating system would adopt the proposed scheme to improve efficiency of process scheduling. In this study, we aim to explore the presentation of the application menu interface for the Apple Watch in operational tasks and completion of the target experience, through market data induction and classification rendering. Subsequently, four application menu interface presentations were output, and experimental tasks were designed. A total of 26 male and female subjects participated in the experiment of using a wearable smartwatch. Subjective measurements and operational performance of the interactive satisfaction scale provided by the subjects were obtained, and a questionnaire for user interaction satisfaction (QUIS) was created. The results show that for the smartwatch application menu interface, regardless of honeycomb-type or list-type menus, adding classification to the presentation type helps users to complete tasks. Notably, the honeycomb-type menu interface achieved a more significant effect. These research results can be used by relevant business owners as a reference and in applications. The treatment of cancer has always been one of the most important fields in the medical field. Due to the safety of radiation therapy, nearly half of the cancers can be treated with radiation. Radiation therapy uses high-energy electromagnetic waves to illuminate cancer cells, so lead must be used to shield the periphery of the treatment room to improve the safety around the treatment environment. However, such shielding will result in difficulties in the transmission of physiological data. In order to solve the above problems, this paper proposes a physiological data monitoring system for the radiation treatment, includes oximeter integration, gateway development and workstation program. The physiological data of the patient during the treatment process can be displayed in the workstation of the control room, and notified when the physiological data is abnormal. In addition, the data recording function is provided as a reference. Through the system development, the safety of cancer patients undergoing radiation therapy is improved. This study provides users with increased interaction with blockchain virtual pet cats through mobile devices and augmented reality, and adds more interactivity and fun. To improve users' experience and communication on blockchain applications, and reduce user's operation burden while allowing more people to access blockchain-related applications through the spread of filter sharing. Due to the growing need for virtual reality (VR) in entertainment, medical, and education, there is more and more research going to deal with the problem of next-generation wireless technologies for VR. Moreover, most of them are machine learning algorithms and confronting challenges to collect datasets from multi-user VR applications. In this paper, we propose a hybrid scenario generator based on a real game, providing interaction adjustable scenarios to support the research which wants to solve the multi-user resource allocation problem. The result shows that the scenarios are more comprehensive than a random distribution. In recent years, with the maturity of virtual reality (VR) hardware devices, many VR software contents have gradually enriched. Existing VR services require level hardware configuration. In the near future, 5G will be able to provide high-bandwidth, low-latency wireless network services, and SDN technology will virtualized mobile network hardware functions, it will be able to create exclusive network transmission services according to different usage scenarios. 5G and SDN are considered the basis for high-quality mobile VR streaming applications. In this study, we developed a VR streaming server prototype exploring the feasibility of managing bandwidth dynamically by SDN functions. In this paper, we propose a music conversion method using deep learning technique to improve the generation of Chinese Guzheng music. Based on human perceptual evaluation, the average score is up to 4.3 on a 5.0 full scale, which represents the similarity between the generated Guzheng music by our method and the real music. It shows that the generated Chinese Guzheng music is able to preserve the feature of the real play. Moreover, this method provides a friendly way for users who have never learned any instrument to generate any desired Guzheng music. Identifying the type of a camera used to capture an investigated image is a useful image forensic tool, which usually employs machine learning or deep learning techniques to train the source camera models. In this research, we propose a forensic scheme to detect and even locate image manipulations based on deep-learning-based camera model identification. Because of the diversity of image tampering, it's difficult to collect a sufficient amount of tampered images for supervised learning. The proposed method avoids preparing tampered images as the training data but chooses to examine the information of original pictures only. We first train a convolutional neural network to acquire generic features for identifying camera models. Next, the similarity measurement using the Siamese network to evaluate the consistency of image block pairs is used to locate tampered areas. Finally, we refine more accurate tampered areas through a refined segmentation network. The contributions of this research include: (1) extending the study of determining image region consistency to forensics applications, (2) designing a better block comparison algorithm, and (3) improving the accuracy of tampered regions. The proposed scheme is tested by public-available tempered image datasets and our own data to verify the feasibility. In this paper, a deep learning-based violin action recognition is proposed. By fusing the sensing signals from depth camera modality and inertial sensor modalities, violin bowing actions can be recognized by the proposed deep learning scheme. The actions performed by a violinist are captured by a depth camera, and recorded by wearable sensors on the forearm of a violinist. In the proposed system, 3D convolution neural network (3D-CNN) and long short-term memory (LSTM) deep learning algorithms are adopted to generate the action models from depth camera modality and inertial sensor modalities. The features and models obtained from multi-modalities are used to classify different violin bowing actions. A fusion process from different modalities can achieve higher recognition accuracy. In this paper, we generate a violin bowing actions dataset for preliminary study and system performance evaluation. In this work, we propose a transfer learning pipeline for gender and age prediction using images from IMDB-WIKI dataset. Firstly, we freeze all layers in pre-trained ImageNet models. Then, the models are trained for four stages with scheduled learning rates and the blocks of layers are unlocked consecutively in accordance to the schedule. We apply multi-output neural network paradigm to predict age and gender simultaneously and the final loss function is based on the combination of age and gender losses. In our approach, the model has better performance than that of the non-pre-trained model because the later stages of our models reuse features extracted from the pre-trained early stages. At present, convolutional neural networks have good performance while performing the object recognition tasks, but it relies on GPUs to solve a large number of complex operations. Therefore, the hardware accelerator of the neural network has become a central topic in the hardware researchers. This letter presents the design of an FPGA-based neural network accelerator implemented on the Xilinx Zynq-7020 FPGA. We use the binary LeNet model to achieve 91% accuracy in the MNIST dataset and use binary AlexNet model to achieve 67% accuracy in the CIFAR-10 dataset. Meanwhile the hardware resource is only about 10% usage on FPGA of the original design. Recently, fitness and health promotion by doing regular jogging exercises are becoming popular. Having a stable and steady stride frequency is critical for the performance of jogging. This paper reports the progress of the design and implementation of a wearable appliance, namely the Intelligent Socks, for assisting the runner to keep steady stride frequency. The proposed appliance uses a set of pressure sensors to improve the accuracy of stride detection. Moreover, the socks provide vibration and sound feedback to help the runner to synchronize with the reference stride frequency. We have implemented an initial prototype and have conducted experiments to find the best way of providing vibration feedback. Sound can affect human emotions, behavior, and decisions. Furthermore, it can change human perception of direction, distance, size of physical objects, and many more. In this study, experiments were conducted to analyze the influence of Interactive Audio in a virtual reality (VR) environment on human perception of temperature. By experiments, we simulate camping scenes with some campfire in VR environments. The subjects wear VR head-mounted displays, headphones, haptic gloves covered with heating pads and an HTC VIVE mobile locator to feel the campfire temperature. The gloves warm up coupled with a computer-simulated fire sound if the subject's hand reaches the virtual fire. The subjects then categorized the fire's temperature from the experiment. Based on the experiment results, the effect of sound on human perception of temperature can be analyzed. Pitch perception training is one of the important aspects of music learning. Long-term training is required for improving one's sense of pitch. However, conventional standard pitch perception training is typically monotonous and is therefore prone to be unfavored by students. With advancements in multimedia technology, it is currently possible to utilize digital games in pitch perception training. In recent years, virtual reality has become a technology trend with wide application in various fields. In this paper, a pitch perception training game combining the elements of shooting game in VR environment is proposed. In the proposed game, different monsters correspondingly appear at different game levels along with a specific pitch. The player has to use a gun to select the pitch corresponding to the targeted monster using a piano-keyboard-like bullet magazine, and pull the trigger to defeat the monster. Through this game, it is expected that the player will become more motivated to participate in pitch perception training. Investigating Consumer Behavioral Intention in smart technology context Li-Wen Chuang and Shu-Ping Chiu (Lingnan Normal University, Taiwan); He-Wei Tian and Luo-Si Wang (Lingnan Normal University, China) Today, a growing number of trades are improving the internet and smart technology as the main tool to promote their products and services. Increasing the skill and ability both to attract the new consumers of young and elder in smart business context is an important challenge for trade managers and has described their considerable attention from the academic groups. More recently, driven by advance developments in Information and Communications Technologies (ICTs) and Virtual-Reality, Augmented-Reality, and Mixed-Reality technologies, embodied devices , sensors, mobile terminal device ,microchip design , power efficiency and more broadly the Internet of Things (IoT), the chance to remotely connect to physical and virtual objects, mobile devices and products would have given rise to the appearance of smart technologies and smart services. Therefore, these businesses directors imminently want to realize the connections with consumer motivation, trust, security risk and consumer behavioral intention requirement of consumer's using smart technologies, so as to take the crucial chance in future smart business development to find the potential consumers, The main concern of this article would be to understand the need and examine the influences of perceptions on perceived value, reputation, trust and consumer behavioral intention to use these smart technologies. Next, the questionnaire was completed by online survey data (N = 92) collected from active consumers who have participated in using smart technologies. Thus, analyzed with structural equation modeling (SEM), and confirmatory factor analysis (CFA) was also applied, using the software of SmartPLS 2.0, to verify if this empirical questionnaire data would become proper to the proposed model. Lastly, the outcomes of this study would apply the empirical research proof that perceived value, reputation and trust all could affect these consumers behavioral intention to use smart technologies. All these consequences of this article with implications for theory and practice would be further cared enough to keep in mind, too. Herbal tea could maintain our health and handle ill health before symptoms appear, and its usage is easier than taking traditional medicines. Depending on individual characteristics, suitable tea should be different for each individual. To provide effective herbal tea for each one, it is desirable to analyze daily personal behaviors. Combinations of herbs could vary if individual characteristics, effective kinds and amounts are considered. It would be very difficult without very skillful and experienced expertises. We aimed to develop an electrical household equipment even without such expertises. In this study, we would like to introduced our system providing suitable herbal tea for each individual, and case studies of effects of the tea proposed by our system. In recent years, with the continuous development of the sharing economy, people have higher requirements on the reliability and accuracy of identification. In this paper, combined with pervasive computing and other communication perception technologies, a lowcost, efficient and convenient identification system for the Internet of things is designed to replace traditional door locks, improve the security index of access control, realize intelligent unlocking and record user data, so that people's lives become intelligent, professional and integrated. The limited battery capacity of sensor has become the main factor which restricts the lifetime of the whole wireless sensor networks. To solve this problem, prior schemes mainly focus on reducing the travel cost of mobile chargers and have low charging performances. In this paper, we put forward partial collaborative mobile charging scheme based on charging curve (PCLCharge) to reduce time spent in the constant voltage mode to charge more nodes and improve usage of energy. Simulation results show that the proposed algorithm is able to maximize the number of charged nodes and improve the usage of energy. This paper proposes an augmented reality (AR)-based question-answering system (QAS) for wedding dress, which applies mobile application to wedding market. Techniques of AR and QAS are proposed to increment users' satisfaction of selecting and buying wedding dress. In experiments, the results show that the accuracy of the proposed QAS is 97% and higher than other methods and systems. This study proposes a machine learning method based on stream homomorphic encryption computing for improving security and reducing computational time. A case study of mobile positioning based on k nearest neighbors (kNN) is selected to evaluate the proposed method. The results showed the proposed method can save computational resources than others. In this paper, the proposed YOLOv3-tiny-dilated network is presented. The well-known YOLO series neural network is famous for its rapidity, and the lightweight version is more suitable for mobile devices or embedded systems. However, the accuracy of one-stage network is slightly lower. Therefore, scale variation problem is investigated in this paper and we replaced the original Feature Pyramid Network (FPN) of the original YOLOv3-tiny network by three dilated convolution branches of different dilation rates to increase the accuracy of the network. As a result, the mean average precision (mAP) was improved by about 2% in the COCO dataset test. Besides, the proposed network was used to identify driving distraction behavior and obtain good accuracy of 95.47%. Toward the realization of dependable high-end consumer electronics products, it is indispensable to design an efficient fault-tolerant packet routing method to enhance the communication performance of Network-on-Chip (NoC)-based many core processors. In the recent study of fault-tolerant routing, a novel approach has been proposed [1]; packets can pass through faulty nodes, rather than detour them as almost all existing methods do. However, disadvantage of the existing method (called passage-y in this paper) is that routing paths are fixed deterministically once the source and destination nodes of the packets are decided. This will result in large communication latency due to the incapability of avoiding congestion. In this paper, we propose a fault-tolerant adaptive routing method which combines two existing methods, well known west-first and the novel passage-y. Simulation result reveals that our method reduces average communication latency by about 21.5%, compared with the previous passage-y method. Residue Number Systems can solve four major problems of neural network including acceleration, power consumption, area overhead and fault tolerance. However, the most recent work remains two major issues: (1) sign detection or magnitude comparison, and (2) long right shifts. In this paper, we develop a systematic approach to design a low power, compact, fast and reliable neural network automatically. Lossless image compression consists of two processes, pixel prediction and residue encoding. Recently, deep neural network (DNN) has been widely adopted in signal prediction. In this work, we apply a DNN architecture together with the information of surrounding pixels weighted by their similarities for pixel prediction. Simulations show that, with the proposed algorithm, the pixels can be predicted more precisely, which is very helpful for lossless image compression. In recent years, there have been frequent rains and typhoons in Taiwan. Continuous heavy rains can cause river rushed and soared. In addition, terrain problems have caused floods to occur frequently, which will cause flooding, landslides, destroying many facilities and many casualties. In this paper, we proposed a remote video-based river monitoring system that uses some traditional image recognition techniques to obtain river regions. In addition, we also use a deep learning method to find water level to ultimately assess river conditions. The experimental results show that the recall of the water level marker is 94.5%. Since the applications of mobile WiFi and Bluetooth signals are very popular and mature, in this paper, we propose an intelligent parking management system based on the technologies of WiFi positioning and Bluetooth control to manage the vehicles entering or exiting the parking lot. Only a dedicated app for the parking lot needs to be installed on the user's mobile phone, when the user drives near the parking lot, the roadside unit (RSU) end of parking lot can automatically complete the operation of vehicle identification. Based on the convenience and practicability, therefore, such a parking system using WiFi positioning and Bluetooth control is worth popularizing. This paper presents a deep learning-based method for effective image companding. The autoencoder inherits the effectiveness of Convolutional Neural Networks (CNN) and residual learning framework to transform High Dynamic Range (HDR) images to Low Dynamic Range (LDR) and its reverse process. Since, the image companding task involves the non-differentiable operation, thus the encoder and decoder networks are alternately trained using an iterative approach. The experimental result clearly reveals that the proposed method yields a promising result. This paper presents a new technique for converting the Multiple Secret Sharing (MSS) from noise-like form into more friendly appearance. This technique simply utilizes the Chinese Remainder Theorem (CRT) and eXclusive-OR (XOR) operations on generating a set of shared images and recovering a set of secret image. As reported in the Experimental Section, the proposed method performs well in MSS system with friendly constraint. In addition, the proposed method requires simple operation for MSS task. This paper presents an end-to-end rapid prototyping methodology that performs automated and efficient mapping of desired neural networks onto FPGA. The design automation agent is considered as Autobot. An early prototype with the hardware decoder generated on the FPGA has been built, and its functionality has been evaluated. The experimental results show that Autobot can offer rapid end-toend prototyping for neural network hardware generation for proactive BMI control. The correct trace identification of the ionosphere in ionograms is one of the fundamental steps to the automatic scaling of ionograms. In this paper, a new method based on deep learning network is proposed to identify the traces of ionograms and ionospheric irregularities effectively. 12000 ionograms from the Chinese Academy of Sciences Digital Ionosonde installed at Huailai, Wuhan, Naning, Ganzi, and Xiamen are utilized to train the deep learning network and satisfying results with 83.9% mAP are obtained. The ionosphere is an essential part of the near-earth environment. Ionosphere classification, especially the irregularities identification via ionograms in real time has significant meaning in the ionospheric research. In this paper, a method based on CNN is proposed to classify the ionograms automatically. To train the CNN, over 20000 ionograms from Chinese Academy of Sciences Digital Ionosonde installed at Huailai, Wuhan, Hengxian, Ganzi and Xiamen are utilized. A good performance over 83% identification accuracy was achieved by the new method. This study explored the operation principles of active clamp flyback converters. A control strategy was adopted to achieve strong noise immunity, and a voltage mode was configured to attain advantages such as simple circuit design. Experiments were performed to verify the feasibility of the proposed active clamp circuit. The experiment results facilitated the completion of an active clamp flyback converter with a switching frequency of 300 kHz, an output power of 65 W, and an output voltage of 19 V. An improved recycling folded cascade (IRFC) operational transconductance amplifier (OTA) is described in this paper. By applying the damping-factor-control (DFC) compensation technique for recycling folded cascade amplifier, the unity-gain frequency and dc gain are improved significantly. The proposed amplifier has been implemented using TSMC 0.18μm CMOS process. According to the simulation results, the proposed amplifier achieved 157.82MHz in unity-gain frequency and 90.19dB in dc gain while driving the same 5.6pF load. This study proposes a noise-reduction method that utilizes a plane microphone array when the target and the noise sources are in a 2-π (sr) space. Outputs from the microphones are regarded as a 3-dimensional (3D) image, and its spectrum is obtained by the 3D fast Fourier transform. Based on the feature that the spatial spectrum of a target signal concentrates on the direct current (DC) component, we can estimate the noise amplitude. By subtracting it from the observed DC amplitude, we can reduce the noise. The results of a computational experiment performed in this study indicate that the proposed method achieved a noise reduction of approximately 20 dB, which is 15 dB larger than that achieved by a conventional delay and sum beamformer. This paper designs a DC-DC converter and rectifier with resonator for wireless power transfer module (WPTM). The proposed boost DC-DC power converter serving as DC-to-AC converter with an embedded DC rectifier to provide the first-stage DC-DC converter, and charge pump support AC-to-DC converter with an input from the DC rectifier. Two differential, cross-coupled rectifiers as the charge pump circuit mitigate the reverse-leakage issue and they are cascaded to double the output voltage. The resonator uses the coil in multi-mode to drive Bluetooth low energy (BLE) and wireless charging functions for power receiving unit (PRU). This paper proposes to use deep transfer learning (DTL) "layer freezing" method to build deep neural network (DNN) models for a target domain with few data on the basis of well-trained DNN models for a source domain with abundant data. Experiments using the DTL method are conducted for building DNN models to predict product surface roughness of wire electrical discharge machining (WEDM). The experimental results show that DTL indeed can help fast build models with high prediction accuracy for the target domain having few data. In recent years, it has become a popular way to extend the lifetime of wireless sensor networks (WSNs) by using mobile charger. The coverage and connectivity of the WSNs are two important factors influencing the lifetime of WSNs, which have not been well considered. An energy recharging mechanism (ERMMS) using coverage and connectivity to decide which sensors to be charged, is proposed to make the WSN network have cumulative maximum coverage. The simulation results show that the proposed method makes the WSNs have more coverage and the charger charging more sensors at the same time, which indicates that the proposed algorithm can maximize the surveillance quality of the WSNs. This paper proposes a manufacturing quality prediction method, called MTF-CLSTM, to integrate the Markov transition field (MTF) model and the convolutional long short-term memory (CLSTM) neural network for wire electrical discharge machining (WEDM). Experiments are conducted to evaluate the proposed method in terms of the mean absolute percentage error (MAPE). Experimental results show that the proposed method outperforms a related method proposed recently. This paper proposes a regulated pulse current driver with spread spectrum technique to lower electromagnetic interference (EMI) effect. A spread spectrum clock generator (SSCG) is used and implemented with triangular wave bias. The results show a 7dBm reduction of the peak power level with a frequency deviation of 10%, demonstrating that the driver effectively reduces peak power and EMI. Virtual reality (VR)/augmented reality (AR) and its applications have attracted significant and increasing attention recently. However, the stringent quality of service (QoS) requirements and better spectral efficiency have posed the challenges such as higher bandwidth, lower latency and better reliability on the VR/AR communication system. This paper proposes a deep-learning-based (DL-based) precoding and feedback method for mitigating the channel interference of multi-users VR/AR environments. That is, our DL-based method uses the VR/ AR channel state information (CSI) to do radio resource allocation for maximizing the millimeter-wave network throughput. Numerical results show that our DL-based design could significantly enhance the throughput. Transmission Line Pulse (TLP) is often applied for ESD devices' characteristics researches. It indicates the devices' triggering-on behaviors, holding performances and thermal run-away results. In normal cases, the thermal run-away data should correlate to Electrostatic Discharge (ESD) [1]. However, the non-correlations between TLP and ESD can be observed in the gate driving circuits [2], silicide devices [3] and poly fuses [4] . In this paper, analyses for the relations between ESD and TLP are taken into discussions for large array device (LAD). Electrostatic Discharge (ESD) performance of large array device (LAD) is a big challenge since the common ESD skill cannot be use. A novel signal control switching (SCS) architecture for adding LAD's ESD robustness is proposed in this paper. A little layout area is increased, but a huge ESD robustness increase can be obtained. In this paper, two common architectures of passive and active capacitive fingerprint sensors are introduced which are implemented by VIS 0.18μm process. The advantages and drawbacks of each technology are also reviewed. In the era of cloud computing with security, how to outsource a trained model to the server but preserve the model privacy is a significant problem for decision tree classification (DTC). In this paper, we aim for constructing an interactive system to realize privacy-preserving delegation for DTC. We focus on a basic structure of DTC, and believe that it can be generalization. The main technique to preserve privacy is to build secret sharing-based protocols for the model owner, user, and server. However, to overcome communication issues between the model owner and user, we achieve compression by using pseudorandom generators. In the past decade smart devices have become a major driving force behind the growth of semiconductor industry. The fierce competition for market shares and profits has rendered frequent release of new generations with fancier designs, more functionalities and better performance. On the other hand, the chips inside these smart devices are designed with lifetime typically spanning a few generations at least. As a result, many smart devices are thrown away with their chips still functioning. Conventional recycling business simply aims at recovering copper, silver, gold, palladium and other materials, and does not take into consideration the potentially functioning chips. We argue that a much better recycling framework should properly classify and bin the functioning chips from recycled smart devices for reuse so that additional profit can be generated and environment can be better protected. In this paper, we propose the concept of integrated circuit recycling, and demonstrate a statistical health assessment method using artificial neuron network (ANN) based search tree along with an optimal price-binning framework with low-cost measurements. Experimental results show that with the simple measurement of Iddq and Vmin, our health assessment can eliminate lifetime overestimation, while flat ANN has that up to 13%. In addition, our pricebinning algorithm can obtain up to extra 51% profit compared with an intuitive maximum-likelihood based approach. This paper presents an important scene detection method based on anomaly detection using a Long Short-Term Memory (LSTM) for baseball highlight generation. In order to deal with multi-view time series features calculated from tweets and videos, we adopt an anomaly detection method using LSTM. LSTM which can maintain a long-term memory is effective for training such features. Introduction of LSTM into important scene detection of baseball videos is the biggest contribution of this paper. Experimental results show high detection performance by our method. In this paper, a new asymmetrically clipped optical orthogonal frequency-division multiplexing (ACO-OFDM) based optical wireless communication system which uses code-division multiple-access (CDMA) technique to satisfy the requirement of the multi-user transmission is proposed. In the proposed system, each user is assigned one of modified maximum length sequence (modified msequence) which is mainly constructed by performing both cyclic-shifted copy and zero-padding operations on the traditional msequence, as the spreading sequence for supporting the (synchronous) downlink transmission of the system. Since drivers acquire a lot of information with eyes, the estimation of the level of visibility (visibility level) on the winter roads contributes toward enhancing traffic safety. Therefore, this paper proposes a method for estimating the visibility level on the winter roads from closed-circuit television (CCTV) images. In the proposed method, the visibility level of each image is estimated based on features acquired via Fourier transform and Convolutional Neural Network (CNN). Specifically, by constructing support vector machines (SVMs) for each feature, the probabilities of the visibility level are calculated. Furthermore, based on comparison of calculated probabilities of SVMs, the proposed method estimates the visibility level accurately. Experimental results show the effectiveness of the proposed method. This paper constructs a 3D multi-target semantic segmentation, detection and tracking framework by using the point cloud data captured from a LiDAR sensor in the outdoor traffic scene. This paper proposes the 3D multi-target detection by using scene segmentation results of PointNet++ to improve the loss of point cloud data when extracting the spatial features of point cloud data after it converting to the voxel space. The spatial features and the scene segmentation results are also represented as a 2D data by bird's-eye view in which the objects does not overlap in the traffic scene. Since the processing time of detection system is much slower than the data acquisition time. The detected target in the point clouds at each time instant are directly tracked by utilizing the local detection with the intersection-overunion ratio in a prediction region. This paper makes a comparison of two types of stabilization control in a remote control system with haptics, By using a haptic interface devise, a user manipulates another haptic interface device at a remote location while watching video. One is the adaptive viscosity control, and the other is the stabilization control by viscosity. By QoE (Quality of Experience) assessment, we clarify which type of control is better than the other type in terms of the operability of haptic interface device. As a result, we demonstrate that the adaptive viscosity control is better than the stabilization control by viscosity when the network delay is short, and the latter is superior to the former when the network delay is long. However, when the network delay is too long, they are almost the same. Recent years, fingerprint recognition has a high market share in biometrics, but humans' hands are often covered with oil and sweat that they secrete, rainwater or dirt, which often affect the accuracy of fingerprint recognition. This work proposed a finger vein identification technology for low response time that can be implemented in a cost-efficient embedded system using the binary robust invariant elementary feature. In this method allows the features matching with binary robust invariant elementary feature and verification with multi-quality assessment process. In experimental results, that the EER performance are 0.13% and 0.69%, using homemade and public (FVUSM) datasets when the data were collected with training and testing. The method is very suitable for real-time finger vein recognition applications. Depthwise separable convolution is useful for building small and lightweight networks. However, the hardware design of depthwise separable convolution unit has not been well studied. With an analysis, we find that many multiplications in depthwise separable convolution can be omitted. Based on this observation, in this paper, we present a novel hardware design to avoid unnecessary multiplications (by using the clock gating technique) for power saving. Experiments on MobileNetV1 model show that the proposed hardware unit can greatly reduce power consumption.
doi:10.1109/icce-taiwan49838.2020.9258230 fatcat:g25vw7mzvradxna2grlzp6kgiq