Search

Deep convolutional neural networks have been widely used for medical image segmentation due to their superiority in feature learning. Although these networks are successful for simple object segmentation tasks, they suffer from two problems for liver and liver tumor segmentation in CT images. One is that convolutional kernels of fixed geometrical structure are unmatched with livers and liver tumors of irregular shapes. The other is that pooling and strided convolutional operations easily lead to the loss of spatial contextual information of images. To address these issues, we propose a deformable encoder-decoder network (DefED-Net) for liver and liver tumor segmentation. The proposed network makes two contributions. The first is that the deformable convolution is used to enhance the feature representation capability of DefED-Net, which can help the network to learn convolution kernels with adaptive spatial structuring information. The second is that we design a Ladder-atrous-spatial-pyramid-pooling module using multi-scale dilation rate (Ladder-ASPP) and apply the module to learn better context information than the atrous spatial pyramid pooling (ASPP) for CT image segmentation. The proposed DefED-Net is evaluated on two public benchmark datasets, the LiTS and the 3DIRCADb. Experiments demonstrate that the DefED-Net has better capability of feature representation as well as provides higher accuracy on liver and liver tumor segmentation than stateof-the art networks. The available code of DefED-Net we propose can be found from https://github.com/SUST-reynole/DefED-Net. Science and Technology Program of Shaanxi Province of China; National Natural Science Foundation of China

Finding a small object (e.g., earbuds, keys or a wallet) in an indoor environment (e.g., in a house or an office) can be frustrating. In this paper, we propose an innovative system, called HyperEar , to localize such an object using only a single smartphone, based on enhanced time-difference-of-arrival (TDoA) measurements over acoustic signals issued from the object. One major challenge is the hardware limitations of a Commercial-Off-The-Shelf (COTS) phone with a short separation between the two microphones and the low sampling rate of such microphones. HyperEar enhances the accuracy of TDoA measurements by virtually increasing distances between microphones through sliding the phone in the air. HyperEar requires no communication for synchronization between the phone and the object and is a low-cost and easy-to-use system. We evaluate the performance of HyperEar via extensive experiments in various indoor conditions and the results demonstrate that, for an object of 7 m away, HyperEar can achieve a mean localization accuracy of about 15 cm when the object in normal indoor environments.

Artificial intelligence is predicted to play a big part in self-learning, industrial automation that will negotiate the bandwidth of structural health and control systems. The industrial structural health and control system based on discrete sensors possesses insufficient spatial coverage of sensing information, while the distributed condition monitoring has been mainly studied at the sensor level, relatively few studies have been conducted at the artificial intelligence level. This paper presents an innovative method for distributed structural health and control systems based on artificial intelligence. The structural condition was divided into regional and local features, the feature extraction and characterization are performed separately. Structural abnormality recognition and risk factor calculation method were proposed by considering the response values and the distribution patterns of both the regional and the local structural behaviours. The test results show that the method can effectively identify the full-scale and local damage of the structure, respectively. Subsequently, structural safety assessment method for long-span structures at kilometres level in view of fully length strain distributions measured by distributed fiber optic sensors were developed. A series of load tests on the long-span structure were carried out. Finite element (FE) model was developed using finite element code, ABAQUS, and an extensive parametric study was conduct to explore the effect of load cases on the structural responses. The differences in the structural response results among load test, structural safety assessment and FE simulation were investigated. It is shown that AI-based self-learning system could offer suitable speed in deployment, reliability in solution and flexibility to adjust in distributed structural health monitoring and control.

In this article, 100-kHz sampling and switching frequency shunt active power filter (SAPF), which is based on a dual-core DSP TMS320F28377D, is designed for the potential market of high-frequency harmonic suppression in more electric aircraft (MEA) or high-speed train (HST). By utilizing the multithreading and multiprocessing method, different computation segments, such as analog-to-digital conversion, recursive discrete Fourier transform (RDFT), and inner loop current and outer loop voltage control, can be rearranged as several parallel operation groups. Compared with the traditional single pipeline structure of the mostly 10–20 kHz switching frequency commercial SAPF, this parallel execution speeds up the signal processing so that the highly increased sampling frequency makes the current control loop reaches wider bandwidth even with simple proportional integral (PI) regulator. Meanwhile, for wideband harmonic suppression, a multiresolution RDFT strategy is proposed to extend the range of the selected harmonic without sacrificing its extraction accuracy. After raising the sampling frequency to 100 kHz, the proposed SAPF with much easier current controller could suppress wider frequency harmonics than other 10–20 kHz SAPF. A 10- $\text {kV}\cdot \text {A}$ prototype with CREE SiC power module CCS050M12CM2 is set up to verify its effectiveness. The result complies with the IEEE standard 519-2014.

Ride comfort and handling performances are known conflicts for off-road vehicles. Recent publications focus on passenger vehicles on class B and class C roads, while, for off-road vehicles, they should be able to run on rougher roads: class D, class E, or class F roads. In this paper, a quarter vehicle model with nonlinear damping is established to analyze the suspension performance of a medium off-road vehicle on the class F road. The ride comfort, road holding, and handling performance of the vehicle are indicated by the weighted root mean square (RMS) value of the vertical acceleration of the sprung mass, suspension travel, and tire deflection. To optimize these objectives, the genetic algorithm (GA), particle swarm optimization (PSO), and a genetic algorithm based on the particle swarm optimization (GA-PSO) are initiated. The efficiency and accuracy of these algorithms are compared to find the best suspension parameters. The effect of the optimized method is validated by the field test result. The ride comfort, road holding, and handling performance are improved by approximately 20%.

For online video service providers, the accurate prediction of video popularity directly impacts their advertisement revenue, bandwidth provisioning policy and copyright procurement decision. Most of previous approaches only utilize data from a single platform (e.g., view history) for prediction. However, such approaches cannot provide satisfactory prediction accuracy, as video popularity may be affected by many influential features dispersed over multiple platforms. In this paper, we focus on the popularity prediction of online movies and propose a prediction framework called DeepFusion to fuse salient features from multiple platforms so as to boost the accuracy of popularity prediction of online movies. For this purpose, we extract influential factors from Douban, which is a leading movie rating website in China, and Youku, which is one of the largest online video service providers in China. Considering the complexity incurred by numerous parameters, we choose to feed these influential factors into deep neural networks for prediction and thus avoid the limitation of traditional predictive models. Compared with previous approaches, our solution can significantly improve the prediction accuracy over 40%. Moreover, even for movies without any historical views, our approach can also well capture their popular trends and overcome the cold-start problem.

The coronavirus disease 2019 (COVID-19) pandemic is spreading worldwide. Considering the limited clinicians and resources and the evidence that computed tomography (CT) analysis can achieve comparable sensitivity, specificity, and accuracy with reverse-transcription polymerase chain reaction, the automatic segmentation of lung infection from CT scans supplies a rapid and effective strategy for COVID-19 diagnosis, treatment, and follow-up. It is challenging because the infection appearance has high intraclass variation and interclass indistinction in CT slices. Therefore, a new context-aware neural network is proposed for lung infection segmentation. Specifically, the autofocus and panorama modules are designed for extracting fine details and semantic knowledge and capturing the long-range dependencies of the context from both peer level and cross level. Also, a novel structure consistency rectification is proposed for calibration by depicting the structural relationship between foreground and background. Experimental results on multiclass and single-class COVID-19 CT images demonstrate the effectiveness of our work. In particular, our method obtains the mean intersection over union (mIoU) score of 64.8%, 65.2%, and 73.8% on three benchmark datasets for COVID-19 infection segmentation.

Although hyperspectral data provide rich feature information and are widely used in other fields, the data are still scarce. Training small sample data classification is still a major challenge for HSI classification based on deep learning. Recently, the method of mining sample relationships has been proved to be an effective method for training small samples. However, this strategy requires high computational power, which will increase the difficulty of network model training. This paper proposes a modified depthwise separable relational network to deeply capture the similarity between samples. In addition, in order to effectively mine the similarity between samples, the feature vectors of support samples and query samples are symmetrically spliced. According to the metric distance between symmetrical structures, the dependence of the model on samples can be effectively reduced. Firstly, in order to improve the training efficiency of the model, depthwise separable convolution is introduced to reduce the computational cost of the model. Secondly, the Leaky-ReLU function effectively activates all neurons in each layer of neural network to improve the training efficiency of the model. Finally, the cosine annealing learning rate adjustment strategy is introduced to avoid the model falling into the local optimal solution and enhance the robustness of the model. The experimental results on two widely used hyperspectral remote sensing image data sets (Pavia University and Kennedy Space Center) show that compared with seven other advanced classification methods, the proposed method achieves better classification accuracy under the condition of limited training samples.

For the original generative adversarial networks (GANs) model, there are three problems that (1) the generator is not robust to the input random noise; (2) the discriminating ability of discriminator gradually reduces in the later stage of training; and (3) it is difficult to reach at the theoretical Nash equilibrium point in the process of training. To solve the above problems, in this paper, a GANs model with denoising penalty and sample augmentation is proposed. In this model, a denoising constraint is firstly designed as the penalty term of the generator, which minimizes the F-norm between the input noise and the encoding of the image generated by the corresponding perturbed noise, respectively. The generator is forced to learn more robust invariant characteristics. Secondly, we put forward a sample augmentation discriminator to improve the ability of discriminator, which is trained by mixing the generated and real images as training samples. Thirdly, in order to achieve the theoretical optimization as far as possible, our model combines denoising penalty and sample augmentation discriminator. Then, denoising penalty and sample augmentation discriminator are applied to five different GANs models whose loss functions include the original GANs, Hinge and least squares loss. Finally, experimental results on the LSUN and CelebA datasets show that our proposed method can help the baseline models improve the quality of generated images.

The modular multilevel converter (MMC) is attractive for medium- and high-power applications because of its high modularity, availability and power quality. In this study, a quadruple sampling phase-shifted-carrier pulse-width modulation (PSC-PWM) strategy based on hierarchical distributed control architecture for MMC in medium-voltage power distribution static synchronous compensator (DSTATCOM) application is presented. Compared to centralised control architecture, the multi-digital signal processors master-slave hierarchical control architecture can enhance the scalability of MMC-DSTATCOM and the sub-module (SM) number can expand without limitation. To take advantage of the proposed hierarchical architecture, the multiple sampling PSC-PWM technique is employed to improve the converter performance. The spectrum of the electromotive force of MMC with quadruple sampling is almost the same as that of natural sampling, which leads to fewer harmonics in reactive and harmonic current compensating. Besides, it also raises the sampling frequency so that the MMC-DSTATCOM can compensate harmonic at the low SM switching frequency. The effectivenesses of the proposed architecture and modulation method are verified by experimental results obtained from a 60 V/2 kVA downscaled prototype.

Most of the approaches used for Landslide inventory mapping (LIM) rely on traditional feature extraction and unsupervised classification algorithms. However, it is difficult to use these approaches to detect landslide areas because of the complexity and spatial uncertainty of landslides. In this letter, we propose a novel approach based on a fully convolutional network within pyramid pooling (FCN-PP) for LIM. The proposed approach has three advantages. First, this approach is automatic and insensitive to noise because multivariate morphological reconstruction is used for image preprocessing. Second, it is able to take into account features from multiple convolutional layers and explore efficiently the context of images, which leads to a good tradeoff between wider receptive field and the use of context. Finally, the selected PP module addresses the drawback of global pooling employed by convolutional neural network, FCN, and U-Net, and, thus, provides better feature maps for landslide areas. Experimental results show that the proposed FCN-PP is effective for LIM, and it outperforms the state-of-the-art approaches in terms of five metrics, $Precision$ , $Recall$ , $Overall~Error$ , $F$ - $score$ , and $Accuracy$ .

Cooperative driving at signal-free intersections, which aims to improve driving safety and efficiency for connected and automated vehicles, has attracted increasing interest in recent years. However, existing cooperative driving strategies either suffer from computational complexity or cannot guarantee global optimality. To fill this research gap, this paper proposes an optimal and computationally efficient cooperative driving strategy with the polynomial-time complexity. By modeling the conflict relations among the vehicles, the solution space of the cooperative driving problem is completely represented by a newly designed small-size state space. Then, based on dynamic programming, the globally optimal solution can be searched inside the state space efficiently. It is proved that the proposed strategy can reduce the time complexity of computation from exponential to a small-degree polynomial. Simulation results further demonstrate that the proposed strategy can obtain the globally optimal solution within a limited computation time under various traffic demand settings.

The development of artificial intelligence and automatic driving is in full swing, and the application of artificial intelligence technology in automatic driving technology is becoming more and more in-depth. Therefore, this paper proposes a visual automatic driving decision system based on deep learning, which can make automatic driving vehicles keep straight in the lane.

For semantic segmentation of high-resolution remote-sensing images, digital surface models (DSMs) information is useful for improving the accuracy and robustness of the segmentation models. However, since the feature distributions of spectral and DSM images vary significantly in different scenes, it is difficult to fuse them effectively in popular deep network models. To solve this issue, we propose an attention-based DSM fusion network (ADF-Net) for high-resolution remote-sensing image semantic segmentation. The proposed network makes two contributions. The first is that we design an attention-based feature fusion module, which can selectively gather features from spectral and DSM information by channel attention mechanism, and further combine them to get high-quality fusion features. The second is that we introduce a residual feature refinement module to reduce the redundant information from skip connection adaptively. We evaluate the proposed network on the ISPRS Vaihingen and Potsdam datasets, experimental results demonstrate that our model outperforms state-of-the-art methods.

In order to solve the above problems, this paper proposes a method of interest points recommendation based on location-based social network fusion of multiple situation information based on the user data of check-in on location-based social network. The social network is transformed into a friend set, and the user interest information model is established based on the user and interest point backward lookup table method, which reduces the running time of interest point recommendation and improves the recommendation efficiency. LDA model is used to model the user’s emotion, and the topic probability distribution of user’s comments is calculated, and the similarity of user’s emotional tendency is mined out. The geographic impact information is analyzed, the geographic impact information model is established, and the access probability of geographic information is calculated. The results show that the proposed method can improve the accuracy of recommendation, reduce the impact of data sparsity, and improve the efficiency of recommendation.

The recursive discrete Fourier transform (RDFT) can extract the frequency components of periodic signals quickly and accurately, which makes the shunt active power filter (SAPF) capable of selective harmonic suppression. However, with the increase of SAPF switching frequency, digital signal processor (DSP) needs to complete 6l±1(l is a positive integer) characteristic harmonic current operation within 50th in a shorter switching cycle. When the switching frequency reaches 100kHz, DSP cannot complete all specified harmonic operations in each switching cycle. In order to make DSP complete as many harmonic suppression as possible, the influence of different sampling frequency on the calculation of harmonic suppression instructions is analyzed firstly, and the minimum allowable sampling frequency is determined. Then, the RDFT harmonic instructions in DSP program are processed by frequency division extraction operation to shorten the calculation amount of harmonic instructions in each switching cycle of DSP, so more specified harmonic suppression is realized. Finally, the strategy is verified on an SAPF with a switching frequency of 100kHz and a capacity of 10KVA. The experimental results show that the strategy improves the suppression ability of high-frequency SAPF significantly.

This design is based on the FPGA platform and builds an automatic number plate recognition system with an on-chip Cortex-M3 core, and collects the number plate photo through the OV5640 camera, and converts it into a digital signal for the image processing module. The image processing module can obtain a clear binary image after gray-scale processing, binarization processing and median filtering. Then it determines the position and character information of the number through vertical and horizontal projection and numerical recognition. The numerical result obtained by the recognition is input to Cortex-M3 through the AHB bus, and the program can be programmed and debugged using keil software on the PC side, and finally the number plate recognition result is displayed on the LCD. The recognition speed of the number plate recognition system reaches 0.4 seconds per picture, and the average correct rate reaches more than 98.5%, achieving the overall performance goal of the system.

Photonic-based reservoir computing (RC) systems have attracted significant attention. Integrated and purely passive systems are compatible with complementary metal-oxide-semiconductor devices, but are limited by the lack of non-linear components. This study consists of two parts: firstly, a review on the published integrated and passive RC system is presented. The review focuses on the structural configuration (rather than the mathematical model) of the neural network; secondly, a new approach for achieving an integrated and passive photonic RC system is introduced and discussed. This approach employs a mode combiner in front of the reservoir to achieve an extra non-linearity in a purely passive device. Moreover, the approach is numerically investigated, and an XOR (exclusive or) task is used to test the device, and the result shows that the new approach satisfies the requirement of an RC system.

The V-Net based 3D fully convolutional neural networks have been widely used in liver volumetric data segmentation. However, due to the large number of parameters of these networks, 3D FCNs suffer from high computational cost and GPU memory usage. To address these issues, we design a lightweight V-Net (LV-Net) for liver segmentation in this paper. The proposed network makes two contributions. The first is that we design an inverted residual bottleneck block (IRB block) and a 3D average pooling block and apply them to the proposed LV-Net. Compared with vanilla convolution, depth-wise convolution and point-wise convolution employed by the IRB block can not only reduce the number of parameters significantly, but also extract features sufficiently well by decoupling cross-channel corrections and spatial correlations. The second is that the LV-Net employs 3D deep supervision to improve the final loss function in training phase, which makes the proposed LV-Net acquire a more powerful discrimination capability between liver areas and non-liver areas. The proposed LV-Net is evaluated on public LiTS dataset, and experiments demonstrate that the proposed LV-Net is superior to popular 2D and 3D networks in terms of segmentation performance, parameter quantity and computational cost.

Motifs are widely used as textures on daily products for their regular and repetitive aesthetics. However, motif design is hard for novices. In this article, we introduce a design exploring system to enable novices to experience motif design easily by simple interactions. We propose the symmetry dependency syntax to encode motif elements and symmetric relations to diverse tree structures. Based on that, we provide motif variation mechanisms, including propagation, adjustment, and attachment, to generate new motifs while recovering motif symmetries broken by the user's manipulations on a motif. When the user deletes, inserts, transforms or replaces an element in a motif, the system automatically generates diverse new symmetric motifs by an encoding-variation flow. Experimental results show that our system can generate diverse motifs in regular visual effects with simple editing interactions, significantly reduce the barrier of motif design for novices.

Road segmentation for all-day outdoor robot navigation is a difficult problem, for the image quality in some time is considerably terrible. In this paper, we propose an effective method to solve this problem. For an outdoor image in any time, the road segmentation can be separated into two stages. Firstly, a supervised generative network is trained to map the outdoor images in any time to the images with rich information. Secondly, a semantic segmentation network outputs a binary segmentation result. Our main contributions include: (1) firstly implementing road segmentation for all-day outdoor robot navigation with a low cost; (2) constructing a supervised generative network for domain mapping and (3) building a dataset for road segmentation for the outdoor images in any time. Our method is evaluated on three datasets. The results indicate that our method achieves a comparable performance with the state-of-the-art approaches.

Aiming at few-shot classification in the field of hyperspectral remote sensing images, this paper proposes a classification method based on cross-scene adaptive learning. First, based on the unsupervised domain adaptive technology, cross-scene knowledge transfer learning is carried out to reduce the differences between source scene and target scene. At the same time, depthwise over-parameterized convolution is used in the deep embedding model to improve the convergence speed and feature extraction ability. Second, two symmetrical subnetworks are designed in the model to further reduce the differences between source scene and target scene. Then, Manhattan distance is learned in the Manhattan metric space in order to reduce the computational cost of the model. Finally, the weighted K-nearest neighbor is introduced for classification, in which the weighted Manhattan metric distance is assigned to the clustered samples to improve the processing ability to the imbalanced hyperspectral image data. The effectiveness of the proposed algorithm is verified on the Pavia and Indiana hyperspectral dataset. The overall classification accuracy is 90.90% and 65.01%. Compared with six other kinds of hyperspectral image classification methods, the proposed cross-scene method has better classification accuracy.

\(k-\)core is one type of cohesive subgraphs such that every vertex has at least k degree within the graph. It is widely used in many graph mining tasks, including but not limited to community detection, graph visualization and clique finding. Frequently decomposing a dynamic graph to get its \(k-\)cores brings expensive cost since \(k-\)cores evolve as the dynamic graph changes. To address this problem, previous studies proposed several maintenance solutions to update \(k-\)cores based on a single inserted (removed) edge. Unlike previous studies, we maintain affected \(k-\)cores from the sparsest to the densest, so the cost of our method is determined by the largest core number of a graph. Experimental results show that our approach can significantly outperform the previous algorithms up to 3 order of magnitude for real graphs tested.

A personalized collaborative filtering recommendation algorithm based on a linear regression model. Constructing the linear regression model based on the user label weight matrix and the user-item scoring matrix, and the gradient regression method is used to minimize the value of the linear regression cost function to obtain the item label. Then, the user and item label weight matrix are substituted into the linear regression model to obtain the user's predicted scores for all unrated items. Using the SlopeOne algorithm principle, calculate the difference between the predicted score and the actual score, and the predicted result is adjusted to obtain the final predicted score. Sort the results and recommend Top-N items to target users. Experiments show that the algorithm's recommendation accuracy is significantly improved than the traditional collaborative filtering algorithm. And the recommended results are interpretable and can meet the individual needs of users.

In this paper, a unique 100kHz sampling and switching frequency shunt active power filter (SAPF), which is based on a dual-core DSP TMS320F28377D, is designed to achieve high control bandwidth for the potential market of wide frequency range harmonic suppression and resonance damping in future more electric aircraft (MEA) or high-speed railway power grid. By rearranging different computation segments, such as analog-to-digital conversion, recursive discrete Fourier transform (RDFT), inner-loop current and outer-loop voltage control, the traditional single pipeline code structure in DSP can be reformed as two parallel running groups. This multitasking strategy provides the proposed SAPF with huge advantages on higher frequency harmonic suppression compared to the mostly commercial SAPF with 10~20kHz switching frequency. Detailed parameter design of LCL filter, DC-side capacitor and deadtime are also presented. A 10kVA prototype utilizing CREE SiC power module CCS050M12CM2 is set up to verify the effectiveness of the proposed SAPF. The results show that the total harmonic distortion can be limited within 5%, which complies with the IEEE power quality standard 519-2014.

Finding a small object (e.g., keys or a wallet) in an indoor environment (e.g., in a house or an office) can be frustrating. In this paper, we propose an innovative system, called HyperEar, to localize such an object using only one single smartphone, based on enhanced time-difference-of-arrival (TDoA) measurements over acoustic signals issued from the object. One major challenge is the hardware limitations of a Commercial-Off-The-Shelf (COTS) phone with a short separation between the two microphones and the low sampling rate of such microphones. HyperEar enhances the accuracy of TDoA measurements by virtually increasing distances between microphones through sliding the phone in the air. HyperEar requires no communication for synchronization between the phone and the object and is a low-cost and easy-to-use system. We evaluate the performance of HyperEar via extensive experiments in various indoor conditions and the results demonstrate that, for an object of 7m away, HyperEar can achieve a mean localization accuracy of about 15cm when the object in normal indoor environments.

Currently, a cloud database is employed to serve on-line query-intensive applications. It inevitably happens that some cloud data nodes storing hot records are facing high frequent query requests while others are rarely visited or even idle. Therefore, how data are dynamically allocated and migrated at runtime has significant impact on query load distribution and system performance. Existing system adopt centralized approaches, and they face two main challenges: (1) Query load on individual node cannot be always balancing even if the data are fairly distributed; (2) For each node, the dynamic changes of configuration resources cannot be captured during the runtime. To this end, this paper presents an environment-aware market strategy based system, named e-MARS, for reasonable data migration to achieve query load balance in cloud database. In e-MARS, cloud database is modeled as a cloudDB market, while data nodes are regarded as intelligent traders and the query load as commodity. Each trader is aware of its local environmental re-sources, such as computing capacity, disk volume, based on which the trader itself decides how to trade the query load and migrates the corresponding data. In this way the cloudDB market will achieve equilibrium. Experiments are conducted on the real communication data, and e-MARS significantly enhances the efficiency. Compared with HBase Balancer, more than 65% improvement is achieved in terms of query response time.

Domain adaptation for semantic segmentation is a challenging problem for two reasons. One reason is that annotating labels is an extremely high cost work. Another reason is that the domain gap between the source and target domains limits the performance of semantic segmentation. In this paper, we propose an unsupervised knowledge based domain adaptation method for semantic segmentation. The proposed method consists of three steps. First, the common knowledge is loaded from the source and target domains. Then, the loaded knowledge is filtered according to the specific input image. In the end, the filtered knowledge is fused with the high-level features to guide domain adaptation. Our main contributions are: (1) a first novel knowledge based domain adaptation approach for semantic segmentation and (2) a triangular constraint for knowledge loading, in which the semantic vectors are smoothly imported. Experimental results on three datasets indicate that our method achieves competitive results in some scenarios compared with the state-of-the-art approaches.

The importance of high-performance computing (HPC) motivates an increasing number of students to study parallel programming. However, a major obstacle for students to learn parallel programming is the lack of large-scale computing resources and feedback for their programs. In this paper, we design and implement an online HPC educational programming system, which provides free computing resources for students and the support of multiple HPC programming languages. The students can easily write HPC programs on our platform and submit to the Tianhe-2 supercomputer for execution. The execution results will be returned and displayed on the front-end web browser of users. In addition, our system also supports code evaluation and feedback debugging via integrating mpiP and TAU into our system kernel. The evaluation results and feedback allow students to look into the execution details of their programs and further optimize their submitted programs.

It is a challenge to recognize holoscopic 3D (H3D) micro-gesture based on general vision techniques because images captured by H3D imaging system are unclear, i.e., the captured images include a large number of blurred grids. Many feature extraction methods can not be directly used for H3D images because the edge information of the grids will be captured. In this paper, we propose a fast and robust preprocessing method for H3D image reconstruction. The reconstructed images are clear and can be used directly for feature extraction or feature learning. Two contributions are presented in this paper. Firstly, we propose a bi-directional morphological filter used for enhancing the grids in an H3D image. Secondly, we propose a fast clustering algorithm with spatial information to extract grids from the H3D image. Because bi-directional morphological filter is able to incorporate local spatial information to the objective function of the fast clustering algorithm, the grids in H3D images are removed completely. Moreover, because the fast clustering algorithm perform clustering on gray levels of H3D images, a small computational cost is required. The proposed method is used to reconstruct H3D images to obtain multiple images with low resolution captured for 3D gesture recognition. Experiments show that the proposed preprocessing method is not only able to obtain better images that are clear and suitable for feature extraction or feature learning, but also is able to improve recognition accuracy in the micro-gesture recognition based on H3D imaging systems.

Lacking of appropriate learning facilities, the traditional approach to learn high-performance computing (HPC) is commonly theory-oriented without sufficient hands-on programming experiences. To improve the hands-on experiences of HPC learners, we design and implement a flexible and adaptive online HPC learning platform in this paper, called EasyHPC. This platform contains various online course modules such as quiz bank, interactive community, and virtual laboratory. In our system, various HPC theoretical and experimental learning activities can be conducted online, such as assigning HPC parallel programming tasks, and creating HPC questions and collecting students' submissions. It is convenient for students to study HPC-related courses in our platform, submit course assignments, exchange ideas, and complete HPC programming tasks. Our preliminary learning trials have shown that our system can effectively improve the hands-on experience of our students by providing an integrated HPC learning and programming environment. Our students are able to achieve various HPC capstone projects in our platform to develop their system capability.

We report a portable "sample to answer" system for the rapid detection of airborne pathogens for the first time. The system contains a key microfluidic chip which fulfills both pathogen enrichment and biological identification functions. The system realizes simple operation and less human intervention as well as minimum reagent contamination. The operation is user-friendly and suitable for field and point-of-care applications. The system is capable of handling detection of different pathogens by changing the primers.

DBSCAN is a classic density-based clustering technique, which is well known in discovering clusters of arbitrary shapes and handling noise. However, it is very time-consuming in density calculation when facing high dimensional data, which makes it inefficient in many areas, such as multi-document summarization, product recommendation, etc. Therefore, how to efficiently calculate the density on high dimensional data becomes one key issue for DBSCAN-based clustering technique. In this paper, we propose a fast algorithm for DBSCAN-based clustering on high dimensional data, named Dboost. In our algorithm, a ranked retrieval technique adaption named $$WAND^\#$$ is novelly applied to improving the density calculations without accuracy loss, and we further improve this acceleration by reducing the invoking times of $$WAND^\#$$. Experiments were conducted on wire voltage data, Netflix dataset and microblog corpora. The results showed that an acceleration of over 50 times were achieved on wire voltage data and Netflix dataset, and 100 more times can be expected on microblog data.

Microblogs are considered as We-Media information with many real-time opinions. This paper presents a Twitter-OpinMiner system for Twitter sentiment analysis evaluation at SemEval 2015. Our approach stems from two different angles: topic detection for discovering the sentiment distribution on different topics and sentiment analysis based on a variety of features. Moreover, we also implemented intra-sentence discourse relations for polarity identification. We divided the discourse relations into 4 predefined categories, including continuation, contrast, condition, and cause. These relations could facilitate us to eliminate polarity ambiguities in compound sentences where both positive and negative sentiments are appearing. Based on the SemEval 2014 and SemEval 2015 Twitter sentiment analysis task datasets, the experimental results show that the performance of Twitter-OpinMiner could effectively recognize opinionated messages and identify the polarities.

A newly developed non-symmetric sinusoidal membership function (NSSMF) is constructed. An improved fuzzy neural network controller using NSSMF is constructed to control the speed of ultrasonic motors. A dynamic algorithm with adaptive learning rate is used to train FNNC online. The global convergence of the FNNC systems could be guaranteed by adjusting the adaptive learning rate. The validity of the proposed scheme is examined by simulated experiments.