Search

Sparsity of rating data is a severe problem to be solved in modern recommendation researches. The fusion recommendation method is an effective solution for the problem. The method combines rating data and other types of user feedback data, such as reviews and image, to improve performance of the traditional recommendation algorithms. Some researchers have proposed fusion recommendation algorithms based on BP (Back Propagation) neural network and achieved better results. However, some existing fusion recommendation algorithms based on BP neural network still have some shortcomings. They rely on the assistance of the traditional recommendation algorithms. Moreover, the high complexity of the fusion processes of these algorithms possibly has negative impacts on the fusion effects. In this paper, we modify the fusion recommendation algorithm and propose the NNFR (neural networks fusion recommendation) model. This model improves the structure of BP neural network by specially designing the structure of network layers. User reviews and ratings can be processed in two separate sub-networks respectively and further fused in the fusion layer. The fusion features of user reviews and ratings are directly applied to perform recommendation, in order to avoid the assistance of the traditional recommendation algorithms and improve the fusing efficiency and quality. Experimental results indicate that the outstanding performance of NNFR model than comparative recommendation algorithms on rating predictions and top-k recommendations. Moreover, NNFR model can still produce high-quality recommendation results in the scenarios of sparse data.

State-of-the-art deep dehazing models are often difficult in training. Knowledge distillation paves a way to train a student network assisted by a teacher network. However, most knowledge distill methods are used for image classification and segmentation as well as object detection, and few investigate distilling image restoration and use different task for knowledge transfer. In this paper, we propose a knowledge-distill dehazing network which distills image dehazing with the heterogeneous task imitation. In our network, the teacher is an off-the-shelf auto-encoder network and is used for image reconstruction. The dehazing network is trained assisted by the teacher network with the process-oriented learning mechanism. The student network imitates the task of image reconstruction in the teacher network. Moreover, we design a spatial-weighted channel-attention residual block for the student image dehazing network to adaptively learn the content-aware channel level attention and pay more attention to the features for dense hazy regions reconstruction. To evaluate the effectiveness of the proposed method, we compare our method with several state-of-the-art methods on two synthetic and real-world datasets, as well as real hazy images.

Convolutional neural networks have made a prominent progress in low-level image restoration tasks. Moire is a kind of high-frequency and irregular interference stripe that appears on the photosensitive element of digital cameras or scanners. It can bring in unpleasant colorful artifacts on images. In this paper, we propose a deep wavelet network with domain adaptation mechanism for single image demoireing, dubbed AWUDN. The feature mapping is mainly performed in the wavelet domain, which can not only cut down computation complexity, but also reduce information loss. Moreover, considering that the images provided by the challenge organizers have strong self-similarity, the global context block is adopted for the learning of feature dependency in different positions. Finally, we introduce the domain adaptation mechanism to fine-tune the pretrained model for reducing the domain gap between training moire dataset and testing moire dataset. Benefiting from these improvements, the proposed method can achieve superior accuracy on the public testing dataset in the NTIRE 2020 Single Image Demoireing Challenge.

This paper reviews the first AIM challenge on bokeh effect synthesis with the focus on proposed solutions and results. The participating teams were solving a real-world image-to-image mapping problem, where the goal was to map standard narrow-aperture photos to the same photos captured with a shallow depth-of-field by the Canon 70D DSLR camera. In this task, the participants had to restore bokeh effect based on only one single frame without any additional data from other cameras or sensors. The target metric used in this challenge combined fidelity scores (PSNR and SSIM) with solutions' perceptual results measured in a user study. The proposed solutions significantly improved baseline results, defining the state-of-the-art for practical bokeh effect simulation.

Degraded underwater images have seriously affected the exploration of the marine environment, causing inconvenience to the marine military, marine engineering, marine research, and other fields. In this paper, we address the underwater image restoration problem by proposing a multiscale iterative network framework that restores the content and details of the image from coarse to fine. First, a endocer-decoder network based on residual blocks and jump connection is designed, which includes several symmetric convolution layers and deconvolution layers to learn an end-to-end mapping of a degraded underwater image to clear image. Then, a multi-scale iterative training strategy is employed to gradually restore the underwater image. Last, a multi-scale loss function is designed to train the network. The experiment results show that the recovery results of the proposed method have better visual perception and quantitative results.

The Computer Numerical Control (CNC) machine may produce a lot of suspended particles in the working environment that have seriously harmed human health, when working. These suspended particles include particulate matter (PM2.5), suspended particulate matter (PM10), and total volatile organic compound (TVOC). Therefore, to improve the working environment, we adopt the Internet of Things (IoT) technology to CNC machines in this study. The indoor air quality sensors connected to the industrial personal computer (IPC) of a CNC machine are used to detect the concentration of suspended particles. Furthermore, in the CNC machine, we also build a humanized interface to analyze the changes of concentration in the time period, and automatically start other devices to reduce it when detecting the high concentration.

Since haze degrades an image including contrast decreasing and color lost, which has a negative effect on the subsequent object detection and recognition. single image dehazing is a challenging visual task. Most existing dehazing methods are not robust to uneven haze. In this paper, we developed an adaptive distillation network to solve the dehaze problem with non-uniform haze, which does not rely on the physical scattering model. The proposed model consists of two parts: an adaptive distillation module and a multi-scale enhancing module. The adaptive distillation block reassigns the channel feature response via adaptively weighting the input maps. And then the important feature maps are separated from the trivial for further focused learning. After that, a multi-scale enhancing module containing two pyramid downsampling layers is employed to fuse the context features for haze-free images restoration in a coarse-to-fine way. Extensive experimental results on synthetic and real datasets demonstrates that the proposed approach outperforms the state-of-the-arts in both quantitative and qualitative evaluations.

Nowadays, drones are expected to be used in several fields, especially flying indoors to explore the surroundings. In this paper, we present a novel method that uses ArUco markers as a reference to improve the accuracy of a drone on straight takeoff, flying forward, and landing based on Reinforcement Learning (RL). Particularly, the drone first detects a specific marker with its onboard cameras. Then, it calculates the position and orientation relative to the marker so as to adjust its actions for achieving better accuracy with an RL method. We perform several simulation experiments by using the Robot Operating System (ROS) and its Gazebo simulator. Simulation results show that our proposed method can efficiently improve the accuracy of the considered actions.

This paper simulates the unit commitment planning and ancillary services in the day-ahead power market of IEEE power system. The unit commitment of power and ancillary services to safe operation of the system planning for Power, Automatic Generation Control (AGC), Real-time Spinning Reserve (RSR) and Supplemental Reserve (SR). This thesis proposed the application of Taguchi Method (TM) for the discretization rule with Enhanced Particle Swarm Optimization (EPSO) to avoid the equality and Inequality of constraint, which can quickly reach the optimal solution with a better performance and accuracy of unit commitment problem for day-ahead power market. Using TM-EPSO algorithm to solve unit commitment and economic dispatch problem which can quickly reach the optimal solution and accuracy for day-ahead power system.

In recent years, we have witnessed a sudden increase in mosquito-borne diseases and related casualties. This makes it important to have an efficient mosquito classification system. In this paper, we implement a mosquito classification system which is capable of identifying Aedes and Culex (types of the mosquito) automatically. To facilitate the implementation of such Internet of Things (IoT) based system, we first create a trap device with a stable area for filming mosquitoes. Then, we analyze video frames in order to reduce the video size for transmission. We also build a model to identify different types of mosquitoes using deep learning. Later, we fine-tune the edge computing on the trap device to optimize the system efficiency. Finally, we integrate the device and the model into a mosquito classification system and test the system in wild fields in Taiwan. The tests show significant results when the experiments are conducted in the rural area. We are able to achieve an accuracy of 98% for validation data and 90.5% for testing data.

This paper presents an active battery balancing circuit based on optimized flyback converter.Through modularization and hierarchical distributed design,the battery balacncing circuit can be easily expanded for large-scale battery packs. There are two improvements in the balancing circuit based on the flyback convertor. The first advantage of the balancing circuit increase the RCD clamp circuit, reducing the energy consumption,preventing the MOS transistor switch from being reversed by the flyback voltage. The second optimization apply for a predictive gate drive technology which replaces the flyback converter rectifier diode with a MOS transistor. It not only improves the efficiency of the battery balancing circuit,but also increases the adaptive dead-time, improving the system Stability. Finally, an experiment using 6 Lithium-ion batteries proves the feasibility of the balanced scheme proposed in this paper.

This paper reviews the first challenge on image dehazing (restoration of rich details in hazy image) with focus on proposed solutions and results. The challenge had 2 tracks. Track 1 employed the indoor images (using I-HAZE dataset), while Track 2 outdoor images (using O-HAZE dataset). The hazy images have been captured in presence of real haze, generated by professional haze machines. I-HAZE dataset contains 35 scenes that correspond to indoor domestic environments, with objects with different colors and specularities. O-HAZE contains 45 different outdoor scenes depicting the same visual content recorded in haze-free and hazy conditions, under the same illumination parameters. The dehazing process was learnable through provided pairs of haze-free and hazy train images. Each track had ~ 120 registered participants and 21 teams competed in the final testing phase. They gauge the state-of-the-art in image dehazing.

This paper highlights the feasible routes of using resistive memory (RRAM) for accelerating online training of deep neural networks (DNNs). A high degree of asymmetric nonlinearity in analog RRAMs could be tolerated when weight update algorithms are optimized with reduced training noise. Hybrid-weight Net (HW-Net), a modified multilayer perceptron (MLP) algorithm that utilizes hybrid internal analog and external binary weights is also proposed. Highly accurate online training could be realized using simple binary RRAMs that have already been widely developed as digital memory.

In the network and communication generation, smart-phone is becoming a basic equipment for most of the people. As such, IoT (Internet of Things) device controlled by smart-phone is feasible and convenient for users. Therefore, this paper proposes to apply infrared signal to implement an IoT gateway that can enable users to control their home appliances such as television, air conditioner, and refrigerator using portable devices such as smart-phone, tablet-computer, and smart-watch. By doing so, a more intelligent smart-home control style is promising for future smart-home living of people. To put it more concretely, in the end of this paper, videos are provided to demonstrate four commercial home appliances that can be controlled by our implemented FPGA-based IoT gateway.

Network Function Virtualization (NFV) on Software-Defined Networks (SDN) can effectively optimize the allocation of Virtual Network Functions (VNFs) and the routing of network flows simultaneously. Nevertheless, most previous studies on NFV focus on unicast service chains and thereby are not scalable to support a large number of destinations in multicast. On the other hand, the allocation of VNFs has not been supported in the current SDN multicast routing algorithms. In this paper, therefore, we make the first attempt to tackle a new challenging problem for finding a service forest with multiple service trees, where each tree contains multiple VNFs required by each destination. Specifically, we formulate a new optimization, named Service Overlay Forest (SOF), to minimize the total cost of all allocated VNFs and all multicast trees in the forest. We design a new $3\rho_{ST}$-approximation algorithm to solve the problem, where $\rho_{ST}$ denotes the best approximation ratio of the Steiner Tree problem, and the distributed implementation of the algorithm is also presented. Simulation results on real networks for data centers manifest that the proposed algorithm outperforms the existing ones by over 25%. Moreover, the implementation of an experimental SDN with HP OpenFlow switches indicates that SOF can significantly improve the QoE of the Youtube service., Comment: Technical Report

Solar energy has the characteristics of intermittent and unstable, and it results in the negative impact on power quality of the utility grid while the penetration rate of the grid-connected photovoltaic generation system is increased. The energy storage system (ESS) can alleviate the negative impact on power quality of the utility grid so as to increase the penetration rate of the grid-connected photovoltaic generation system. Aqueous Hybrid Ion (AHI) battery contains no heavy metals or toxic chemicals, and it is non-flammable and non-explosive. Besides, AHI battery has the characteristics of high cycle life, safety and sustainability. A 5kW DC-coupling distribution power generation system (DPGS) based on photovoltaic and AHI battery set is presented in this paper. The operation modes of this DPGS are divided into self-consumption, charging/discharging schedule and stand-alone. The maximum power efficiency from the solar cell array to the grid is 97.1%. The maximum power efficiency of the bidirectional DC-DC converter is 97.3% and the maximum power efficiency from the AHI battery set to the grid is 94.5%. Therefore, the power efficiency of this DPGS is very high.

This paper develops the study about parking spaces detection in indoor environments, implementing a wireless sensor networks (WSNs) based on Zigbee technology. With the popularity of digital communications technology, there are many people engaged in wireless sensor network applications. Indoor positioning technology is very important area among wireless sensor network applications. In this paper, the fuzzy model converted the Received Signal Strength Indicator (RSSI) into the distance. We combine fuzzy with Minimum Mean Square Error (MMSE) algorithm estimation for positioning.

In this study, a hexa-band antenna for mobile phone has been developed for low specific absorption rate. The proposed antenna has a simple structure comprising of two folded strips and a rectangular ground plane. The antenna has been designed to cover six bands which are GSM850/900 (824–960 MHz), DCS 1800 (1710–1880 MHz), PCS 1900 (1850–1990 MHz), UMTS (1920–2170 MHz), and ISM 2450 (2400–2485 MHz). The designed antenna only occupies a small area of 50×20 mm2 on the system circuit board. By adjusting the strip length, the SAR is effectively reduced over all of the bands. The antenna is a promising candidate for incorporation into smart phones.

This study using Kano's Model and QFD to develop a simple, efficient, and easy to use personal mobile software assessment shoulder and neck pain, providing the subject's shoulder and neck pain can be self-testing by an application. And through the detection database, expert systems and decision support mechanism, results will be provided by the app charts after detection. The user can cause uncomfortable using chart understanding of factors that can be provided by the system is also recommended for subsequent improvement measures, early detection of hazards in the environment, and by improving the working environment and correct personal habits to reduce neck and back pain incidence.

This paper presents a design of controlled robotic arm with myoelectric and body action signals. The implementation uses the sensed signals, via the signal processing of ARDUINO UNO R3 development board and NUC140VE3CN development board (ARM processor), to control the robotic arm wirelessly. The proposed design can be used in the dangerous operation environment. The users can contactlessly control the robotic arm safely. And it can operate specified action repeatedly and accurately for factory manufacture. The rotative angle of robotic arm controlled by Servomotor is decided by pulse width modulation signal obtained from microcontroller via BlueTooth 4.0 wireless technology. The pulse width modulation signal obtained from microcontroller is decided by the sensors located on the human's arm or sensor glove.

This paper proposes a GA-SVM classification method which is applied to the dynamic evaluation of taekwondo. For classifying a dynamic action, we converted a dynamic action signal to a frequency spectrum signal for analysis. However, the useful features were concentrated in a part of the frequency spectrum, and the useless features led to a decline in accuracy, operation speed, and efficiency of the classifier. Therefore, we propose a classification method that involves using a support vector machine (SVM) with a genetic algorithm (GA) for the dynamic evaluation of taekwondo. The GA determines the useful features, and the useless features are eliminated, thus reducing the dimensionality and improving the accuracy of the classifier. The SVM can solve problems of small, nonlinear, and high-dimension samples efficiently, exhibiting superior performance in the classification of a dynamic action.

This paper presents the design of a 5 GHz g m -boosted transformer cross-coupled current-reused Colpitts voltage-controlled oscillator (VCO). The gm-boosting techniques of transformer cross-coupling is adopted to make the oscillation startup condition much easier with a lower current consumption. This Colpitts VCO is implemented in TSMC 0.18μm 1P6M CMOS process with a chip area of 0.655 mm2. The measured frequency tuning range is from 4.63 to 5.04 GHz, and the phase noise is −116.79 dBc/Hz at 1 MHz offset. The VCO core dissipates 3.74 mW from a 1.8 V supply.

This paper analyzes the current situation of watt-hour meter verification and proposes a design scheme of watt-hour meter verification line based on CAN bus and Ethernet. The verification line adopts modular design and the whole line is made up of three necessary verification links, namely withstand voltage verification, initial verification and repeat verification, achieving automatic connecting and disconnecting, automatic verification. A computer, the assembly line central control terminal,monitors the whole processes dynamically. Actual test results show the line's three links work orderly to make the verification process automatic and effective.

This paper reports our implementation of an experimental system to generate DB-T2 signal based on a commercial FPGA (Field Programmable Gate Array) board with additional instruments. The baseband OFDM (Orthogonal Frequency Division Multiplexing) samples are computed offline via a C program. The OFDM samples are stored in an SD card for the FPGA board to read. The FPGA board converts the digital samples to analog baseband signal to be modulated by an I/Q modulator. The generated DVB-T2 signal can be successfully decoded by a commercial DVB-T2 receiver card, and thus the experimental system is validated.

Deep packet filtering (DPF) has been demonstrated as an essential technique for effective fine-grained access controls, but it is commonly recognized that the technique may invade the individual privacy of the users. Secure computation can address the tradeoff between privacy and DPF functionality, but the current solutions limit the scalability of the network due to the intensive computation overheads and large connection setup delay, especially for the latest network paradigm, network function virtualisation (NFV) and software-defined network (SDN). In this paper, therefore, we propose a privacy-preserving deep packet filtering protocol, named DPF-ET, that can efficiently perform filtering function over encrypted traffic while diminishing the communication overhead and setup delay for the controller in SDN. DPF-ET guarantees the data privacy for users and remains rule privacy for the network owner. The implementation results on an experimental HP SDN/NFV platform demonstrate that the proposed DPF-ET outperforms the current approaches by reducing 250 times in the communications overhead and 32 times in the setup delay.