Search

The brain-computer interface (BCI) of steady-state visual evoked potential (SSVEP) is one of the fundamental ways of human-computer communication. The main challenge is that there may be a nonlinear relationship between different SSVEP in other states. For improving the performance of SSVEP BCI, a novel CNN algorithm model is proposed in this study. Based on the discrete Fourier transform to calculate the signal's power spectral density (PSD), we perform zero-padding in the signal's time domain to improve its performance on the PSD and make it more refined. In this way, the frequency point interval in the PSD of the SSVEP is consistent with the minimum gap between the stimulation frequency. Combining the nonlinear transformation capabilities of CNN in deep learning, a zero-padding frequency domain convolutional neural network (ZPFDCNN) model is proposed. Extensive experiments based on the SSVEP dataset validate the effectiveness of our method. The study verifies that the proposed ZPFDCNN method can improve the effectiveness of the SSVEP-based high-speed BCI ITR. It has massive potential in the application of BCI.

There are still many problems in cross-media annotation and retrieval. The semantic gap between the features of multiple media and the low-level visual features do not carry semantics, and the presence of noise has a great impact. This paper proposes a cross-media semantic learning model that combines Parallel and sharing attention mechanism, which sets semantic graph network topology and node information in the graph neural network. Testing on experimental data show that the network structure proposed in this paper can achieve better accuracy than traditional methods.

Different types of media data are represented by the underlying features of different dimensions and different attributes, resulting in heterogeneity and incomparability; on the other hand, in order to use the underlying feature information extracted from cross-media, additional information must be obtained from different styles. There is a compromise between the difference and the semantic ambiguity obtained only from the bottom layer. These problems lead to traditional feature learning methods not suitable for cross-media analysis. This paper proposes a neural network structure that combines feature extraction and context semantics and introduces a new attention mechanism. Tests on experimental data show that the network structure proposed in this paper can obtain a better BLEU score than traditional methods.

Steam submerged jet condensation is a direct contact condensation widely used in many industries. However, this process may cause damage to related equipment due to pressure oscillation, especially at low steam mass flux. This study synchronously investigated the steam–water interface fluctuation of unstable condensation jet and its corresponding pressure oscillation through a horizontal pipe at low steam mass flux to research the condensation oscillation mechanism. Three typical flow patterns, namely, Chugging, hemispherical bubble oscillation (HBO), and encapsulating bubble oscillation (EBO), were distinguished based on the dynamic behavior of steam–water interface and oscillation pressure characteristics. Only the dynamic pressure in EBO region was periodic among the three flow patterns. In the EBO region, the pressure oscillation frequency increased with the increase of steam mass flux and decreased with the increase of water temperature. A flow pattern regime map for horizontal jet condensation related to steam mass flux and water temperature was presented based on experiments. Moreover, the influence of pipe diameters on low steam mass flux was analyzed. For the flow pattern regime map at low steam mass flux, the critical steam mass fluxes showed an inverse relationship with pipe diameter. An empirical criterion was introduced to determine the flow regime map at low steam mass flux.

In recent years, deep learning-based image extrapolation has achieved remarkable improvements. Image extrapolation utilizes the structural and semantic information from the known area of an image to extrapolate the unknown area. In addition, these extrapolative parts not only maintain the consistency of spatial information and structural information with the known area, but also achieve a clear, beautiful, natural and harmonious visual effect. In view of the shortcomings of traditional image extrapolation methods, this paper proposes an image extrapolation method which is based on perceptual loss and style loss. In the paper, we use the perceptual loss and style loss to restrain the generation of the texture and style of images, which improves the distorted and fuzzy structure generated by traditional methods. The perceptual loss and style loss capture the semantic information and the overall style of the known area respectively, which is helpful for the network to grasp the texture and style of images. The experiments on the Places2 and Paris StreetView dataset show that our approach could produce better results.

Simultaneous localization and mapping (SLAM) is considered to be the basic ability of intelligent mobile robots. In the past few decades, thanks to community’s continuous and in-depth research on SLAM algorithms, the current SLAM algorithms have achieved good performance. But there are still some problems. For example, most SLAM algorithms have the assumption of a static environment, but in real life, most of the environment contains moving objects, so how to deal with the moving objects in the environment requires careful consideration. What’s more, traditional geometric maps cannot specific environmental semantic information for mobile robots, so how to make robots truly understand the surrounding environment to complete some advanced tasks is also a difficult problem. In this paper, we design a scheme to improve the accuracy and robustness of SLAM in a dynamic environment. And we realize the perception of semantic information of objects in the environment through the object detection algorithm of deep learning neural network.

Create a 3D face model from a 2D face image, generally extract facial feature points, calculate a 3D deformation model, and perform deformation and stretching on the generated face database. However, this approach is not only time-consuming but also has no calculation errors. Ideally, neural networks’ use to obtain deformation model parameters is also affected by factors such as pose, angle, and datasets. 3D face reconstruction methods rely excessively on the accuracy of the labeling and the face detector’s accuracy. This article proposes a method that is not affected by pose. We adopt a feature point extractor that can obtain more features, design an hourglass network to get a model, and consider each feature area differently, effectively using the feature point information. Map from two-dimensional coordinates to three-dimensional space to achieve face reconstruction, and obtain a high-precision face model. We do experiments on the three-dimensional face datasets AFLW2000-3D and 300W-3D. The results show that this method can obtain good performance in face multi-angle reconstruction, and the accuracy is also improved.

In automatic navigation robots, robotic autonomous positioning is one of the most difficult challenges. Simultaneous Localization and Mapping (SLAM) technology can incrementally construct a map of the robot’s moving path in an unknown environment while estimating the position of the robot in the map, providing an effective solution for robots to fully navigate autonomously. The camera can obtain corresponding two-dimensional digital images from the real three-dimensional world. These images contain very rich color, texture information and highly recognizable features, which provide indispensable information for robots to understand and recognize the environment based on the ability to autonomously explore the unknown environment. Therefore, more and more researchers use cameras to solve SLAM problems, also known as visual SLAM.

In the automatic navigation robot field, robotic autonomous positioning is one of the most difficult challenges. Simultaneous localization and mapping (SLAM) technology can incrementally construct a map of the robot’s moving path in an unknown environment while estimating the position of the robot in the map, providing an effective solution for robots to fully navigate autonomously. The camera can obtain corresponding two-dimensional digital images from the real three-dimensional world. These images contain very rich colour, texture information, and highly recognizable features, which provide indispensable information for robots to understand and recognize the environment based on the ability to autonomously explore the unknown environment. Therefore, more and more researchers use cameras to solve SLAM problems, also known as visual SLAM. Visual SLAM needs to process a large number of image data collected by the camera, which has high performance requirements for computing hardware, and thus, its application on embedded mobile platforms is greatly limited. This paper presents a parallelization method based on embedded hardware equipped with embedded GPU. Use CUDA, a parallel computing platform, to accelerate the visual front-end processing of the visual SLAM algorithm. Extensive experiments are done to verify the effectiveness of the method. The results show that the presented method effectively improves the operating efficiency of the visual SLAM algorithm and ensures the original accuracy of the algorithm.

The condensation induced water hammer (CIWH) phenomenon may cause serious damage to the pipes and related system, which often occurs during the steam-water direct contact condensation process. In this paper, an experimental investigation was performed to study CIWH phenomenon caused by steam-water direct contact condensation in a horizontal pipe. The entire CIWH process was captured by a high speed video camera and its pressure fluctuation was synchronously measured. Four typical flow patterns were observed during CIWH process: stratification flow, wave flow, slug flow and bubble collapse. Bubble collapse would generate a high pressure peak. Based on different variations of steam-water phase interface and pressure fluctuation signals, three types of CIWH were defined: non-periodic CIWH, periodic CIWH and no CIWH. A CIWH region map was given considering the effect of steam mass flux and water temperature. In periodic CIWH region, the generation frequency of CIWH was found to range from 0.19 Hz to 0.39 Hz, which decreased with the rise of steam mass flux and water temperature. A dimensionless correlation was obtained to predict Strouhal number of CIWH generation frequency. Predicted values corresponded well to the experimental data with the deviation in the range of −16% to +23%.

Submerged steam jet condensation is widely applied in various fields because of its high heat transfer efficiency. In this paper, sonic steam jet condensation through a double-hole nozzle was numerically researched. First, the effect of L/D (pitch to diameter) on steam plumes was investigated. When L/D increased from 1.2 to 2.5, two steam plumes transformed from merging into one steam plume to being independent. Besides, thermal hydraulic parameters along radial directions were analyzed, revealing the existence of expansion and compression waves. Radial distributions of steam volume fraction, temperature and velocity were all largely symmetric about nozzle centerline and there all existed minimum values at nozzle centerline position. Finally, the effect of L/D on thermal hydraulic parameters of nozzle centerline was obtained. There existed peak values for steam volume fraction, temperature and velocity of nozzle centerline at one position. As L/D increased, the peak value decreased and its position moved downstream.

Most traditional curing systems are manually or half-artificial operated that requiring the curers to observe the state of tobacco leaves frequently. A novel intelligent real-time curing control system is developed in this paper by acquiring the optical image of tobacco leaves and extracting the color features and texture features to predict and control the temperature and humidity of the curing barn. The tobacco leaves changes from green to yellow and shrinks gradually, and this changing regulation would enhance the intelligence of tobacco curing system. The proposed neural network is designed to predict the set-point values of the adjustment of dry-bulb temperature, wet-bulb temperature and the changing time, which has eleven inputs include three color features, three texture features, ideal dry-wet temperature, ideal wet-bulb temperature, current stage, stage passing time, tobacco leaves varieties and flag. Some experiments are induced and the experimental results show this proposed approach based on color features and texture features could improve significantly the accuracy than that of the similar method only using color features especially in post-curing process.

Three-dimensional steady simulation was performed to investigate the influence of non-condensable gas on stable condensation jet. Condensation model concerning non-condensable gas was inserted into Ansys to simulate the process of condensation with non-condensable gas. Water, steam and non-condensable gas were chosen as the three phases, and mixture multiphase model was employed. The contour of steam and axial temperature distribution were compared between numerical and experimental result, and good agreements were obtained. Then the effect of non-condensable gas on the contours of steam and air, axial parameters distributions such as velocity, temperature and pressure etc, were investigated as the void fraction of non-condensable gas ranges from 0 to 7%. As the amount of non-condensable gas increases, the penetrating length of steam plume is longer and longer, and the thermodynamic parameters along axial direction showed regular changes at the same time. Besides, it was found that non-condensable gas gathered at the end of steam plume, reaching the maximum void fraction.

Coordinated numerical ensemble experiments with six different state-of-the-art atmosphere models were used to evaluate and quantify the impact of global SST (from reanalysis data) on the early winter Arctic warming during 1982–2014. Two sets of experiments were designed: in the first set (EXP1), OISSTv2 daily sea-ice concentration and SST variations were used as the lower boundary forcing, while in the second set (EXP2) the SST data were replaced by the daily SST climatology. In the results, the multi-model ensemble mean of EXP1 showed a near-surface (~850 hPa) warming trend of 0.4 °C/10 yr, which was 80% of the warming trend in the reanalysis. The simulated warming trend was robust across the six models, with a magnitude of 0.36–0.50 °C/10 yr. The global SST could explain most of the simulated warming trend in EXP1 in the mid and low troposphere over the Arctic, and accounted for 58% of the simulated near-surface warming. The results also suggest that the upper-tropospheric warming (~200 hPa) over the Arctic in the reanalysis is likely not a forced signal; rather, it is caused by natural climate variability. The source regions that can potentially impact the early winter Arctic warming are explored and the limitations of the study are discussed. 摘要 本文使用六个不同的最新大气模式进行了协调数值集合实验, 评估和量化了全球海表面温度 (SST) 对1982–2014年冬季早期北极变暖的影响.本研究设计了两组实验:在第一组 (EXP1) 中, 将OISSTv2逐日变化的海冰密集度和SST数据作为下边界强迫场;在第二组 (EXP2) 中, 将逐日变化的SST数据替换为逐日气候态.结果表明: (1) EXP1的多模式集合总体平均值显示0.4 °C/10年的近地表 (约850 hPa) 升温趋势, 为再分析数据结果中升温趋势的80%. (2) 在这六个模式中, 模拟的变暖趋势均很强, 幅度为0.36–0.50 °C/10年. (3) 全球海表温度可以解释北极对流层中低层EXP1的大部分模拟的变暖趋势, 占再分析数据结果的58%. (4) 再分析数据结果中, 北极上空的对流层上层变暖 (约200 hPa) 不是由强迫信号而可能是由自然气候变率引起的.本文还探索了影响北极初冬变暖的可能源区, 并讨论了该研究的局限性.

In view of the importance of the condensation-induced water hammer in various engineering fields, this study performed an analytical modeling and numerical simulation to obtain the map of the onset of the condensation-induced water hammer. Two linear boundaries were experimentally observed for the transitions between no condensation-induced water hammer and non-periodic condensation-induced water hammer and between non-periodic condensation-induced water hammer and periodic condensation-induced water hammer. The analytical modeling based on the energy balance equation was conducted to derive the onset of the condensation-induced water hammer. The analytical model suggested that the map of the onset of the condensation-induced water hammer could be displayed by two essential non-dimensional parameters, such as modified Jakob number and Froude number. The model indicated that the slope of the boundary of the onset of the condensation-induced water hammer was the product of modified Stanton number, non-dimensional interfacial area concentration, and non-dimensional temperature difference. The model could predict the boundaries between no condensation-induced water hammer and non-periodic condensation-induced water hammer and between non-periodic condensation-induced water hammer and periodic condensation-induced water hammer satisfactorily. The one-dimensional nuclear thermal-hydraulic system analysis code, RELAP5, was used for simulating the onset of the condensation-induced water hammer. The obtained boundary corresponded to the experimentally obtained boundary between non-periodic condensation-induced water hammer and periodic condensation-induced water hammer.

Steam condensation-induced water hammer (CIWH) can cause great pressure oscillation, and the knowledge of CIWH is important to pipes and relevant equipment. In this work, an experiment was conducted to investigate the condensation process of entrapped bubbles for periodic CIWH and its pressure oscillation in a horizontal pipe. Results indicated that CIWH was caused by the condensation of entrapped bubbles, and the initial volume of entrapped bubbles of the periodic CIWH kept nearly unchanged when the test conditions were constant. The initial bubble volume increased with the increases in steam mass flux and water temperature. The pressure oscillation intensity of the CIWH increased with the increase in initial bubble volume but decreased with the increase in water temperature. Furthermore, the change rate of bubble volume gradually increased with the condensation of the entrapped bubbles. The final change rate of bubble volume increased with the increase in initial bubble volume but decreased with the increase in water temperature. Finally, a theoretical correlation, that is, a linear relationship between the pressure oscillation intensity and the final change rate of bubble volume, was proposed to calculate the pressure oscillation intensity of the CIWH. The predicted deviation ranged from −44% to 64%.

Convolutional neural networks (ConvNets or CNNs) have been proven very effective in areas such as image recognition and classification. Especially in the field of image classification, the CNN-based method has achieved excellent performance. Performance is an important indicator for evaluating whether a CNN-based classification method is excellent, so it is important to study which factors affect performance. As we all know, image classification performance is affected by the network structure itself and the size of the data set. In particular, data set size have a significant impact on performance. While for most people, a large number of data set are difficult to obtain. Thus, we consider a question of this approach: How does the size of the data set affect performance? In order to clarify this issue, there are 35 groups experiment performed with 5 times experiment in each group (175 experiments in total). For each k-classification experiment, we do 5 groups by increasing the size of the training set. Observe changes in accuracy to analyze the effect of data set size on difference. For the same CNN-based network, experimental results of average accuracy illustrate that the larger the training set, the higher the test accuracy. However, when the training data set are insufficient, better results can be obtained. Furthermore, in each group experiment, the more categories that are classified, the more obvious the performance change. Results of this paper not only can guide us to do experiments on image classification, but also have important guiding significance for other experiments based on deep learning.

Catalytic nickel was successfully incorporated into nanoscale iron to enhance its dechlorination efficiency for trichloroethylene (TCE), one of the most commonly detected chlorinated organic compounds in groundwater. Ethane was the predominant product. The greatest dechlorination efficiency was achieved at 22 molar percent of nickel. This nanoscale Ni-Fe is poorly ordered and inhomogeneous; iron dissolution occurred whereas nickel was relatively stable during the 24-hr reaction. The morphological characterization provided significant new insights on the mechanism of catalytic hydrodechlorination by bimetallic nanoparticles. TCE degradation and ethane production rates were greatly affected by environmental parameters such as solution pH, temperature and common groundwater ions. Both rate constants decreased and then increased over the pH range of 6.5 to 8.0, with the minimum value occurring at pH 7.5. TCE degradation rate constant showed an increasing trend over the temperature range of 10 to 25°C. However, ethane production rate constant increased and then decreased over the range, with the maximum value occurring at 20°C. Most salts in the solution appeared to enhance the reaction in the first half hour but overall they displayed an inhibitory effect. Combined ions showed a similar effect as individual salts.

Genetic network programming (GNP) is a new evolutionary algorithm using the directed graph as its chromosome. A GNP-based rule accumulation (GNP-RA) method was proposed previously for multiagent control. However, in changing environments where new situations appear frequently, the old rules in the rule pool become incompetent for guiding the agent's actions, and therefore updating them becomes necessary. This paper proposes a more robust rule-based model which can adapt to the environment changes. In order to realize this, Sarsa-learning is used as a tool to update the rules to cope with the unexperienced situations in new environments. Furthermore, Sarsa-learning helps to generate better rules by selecting really important judgments and actions during training. In addition, the e-greedy policy of Sarsa enables GNP-RA to explore the solutions space sufficiently, generating more rules. Simulations on the tile world problem show that the proposed method outperforms the previous ones, namely GP and reinforcement learning. © 2014 Institute of Electrical Engineers of Japan. Published by John Wiley & Sons, Inc.

This study is aimed to prevent the agglomeration of Pd/Fe bimetallic nanoparticles and thus improve the efficiency toward degradation and dechlorination of chlorinated organic contaminants. A mesoporous silica with a primary pore diameter of 8.3 nm and a specific surface area of 688 m2/g was prepared and used as the host of Pd/Fe nanoparticles. The Pd/Fe nanoparticles were deposited onto or into the mesoporous silica by reduction of ferrous ion and hexachloropalladate ion in aqueous phase. Batch degradation and dechlorination reactions of trichloroethylene were conducted with initial trichloroethylene concentration of 23.7 mg/L, iron loading of 203 or 1.91 × 103 mg/L and silica loading of 8.10 g/L at 25 °C. Concentration of trichloroethylene occurs on the supported Pd/Fe nanoparticles, with trichloroethylene degrading to 56% and 59% in 30 min on the supported Pd/Fe nanoparticles with weight percentage of palladium to iron at 0.075% and 0.10% respectively. The supported Pd/Fe nanoparticles exhibit better dechlorination activity. When the supported Pd/Fe nanoparticles with a weight percentage of palladium to iron of 0.10% were loaded much less than the bare counterpart, the yield of ethylene plus ethane in 10 h on them was comparable, i.e., 19% vs. 21%. This study offers a future approach to efficiently combine the reactivity of supported Pd/Fe nanoparticles and the adsorption ability of mesoporous silica.

To overcome the performance degradation in the presence of steering vector mismatches, strict restrictions on the number of available snapshots, and numerous interferences, a novel beamforming approach based on nonlinear least-square support vector regression machine (LS-SVR) is derived in this paper. In this approach, the conventional linearly constrained minimum variance cost function used by minimum variance distortionless response (MVDR) beamformer is replaced by a squared-loss function to increase robustness in complex scenarios and provide additional control over the sidelobe level. Gaussian kernels are also used to obtain better generalization capacity. This novel approach has two highlights, one is a recursive regression procedure to estimate the weight vectors on real-time, the other is a sparse model with novelty criterion to reduce the final size of the beamformer. The analysis and simulation tests show that the proposed approach offers better noise suppression capability and achieve near optimal signal-to-interference-and-noise ratio (SINR) with a low computational burden, as compared to other recently proposed robust beamforming techniques.