Search

With the drastic growth of the energy internet in recent years, the energy router, an important energy hub, has seen wide application and unprecedented value due to its many functions such as energy transmission and coordination management. However, the energy router's ability to cope with the grid voltage sag is critical to its stable operation. This paper presents an advanced low voltage ride-through (LVRT) control strategy for the energy router when the symmetrical grid voltage sag occurs. The presented LVRT strategy can assist in voltage recovery and active power maintenance. When detecting a voltage sag, the inner loop regulator is directly controlled which contributes to the amelioration of the response speed as the outer loop regulator is disconnected. Voltage sags of different severities require different reactive and active power injections to support the voltage recovery and reduce the impact on the energy router's power delivery. Simulation results confirm the effectiveness of the presented LVRT strategy.

The hyperspectral unmixing (HU) aims at estimating the spectral signatures of endmembers (or materials) and their corresponding abundance maps of the hyperspectral image (HSI). In this work, we treat the HSI as an 3D cube and propose a new tensor-based HU method. We decompose an HSI data as the sum of several multilinear rank-(Lr, Lr, 1) terms (or LL1 model). Based on the fact that the latent factors of LL1 model are physical meaningful (i.e., abundance maps and spectral signatures), we build a nonnegative tensor decomposition optimization model with the low-rank constraint and impose an implicit regularizer to exploit the nonlocal self-similarity prior of abundance maps-whose related subproblem can be easily solved under the plug-and-play framework. We develop an alternating direction method of multipliers algorithm to solve the proposed model. Numerical experiments demonstrate the effectiveness of our algorithm.

In recent year, due to motility and wide coverage, unmanned aerial vehicle (UAV) has been widely applied in surveillance system. Human action recognition in UAV video is essential for surveillance video understanding. However, existing action recognition methods suffer from heavy computing, which makes it hard to deploy in real applications. In this paper, a lightweight action recognition method for UAV video(LARMUV) is proposed. This method is based on TSN and adopt Mo-bileNetV3 as backbone, which greatly reduces amount of computing and parameters. Self-attention mechanism is adopted to capture temporal structure among different frames. For loss function, Focal Loss is used to putting more focus on hard, misclassified examples. Last but not least, knowledge distillation is employed to enhance the performance of our model, which transfer knowledge from a larger teacher model to student model. Experimental results on HMDB51, UCF101 and UAV dataset show that our method can achieve competitive performance compared to baseline methods while run in real-time mode.

The underlying cryptography suite of Hyperledger Fabric only supports international cryptographic algorithm, and does not support Chinese commercial cryptographic algorithm SM2/3/4, which prevents the promoting of Hyperledger Fabric in China. To further promote the application of Hyperledger Fabric in China, we retrofit the underlying cryptography suite of Fabric v1.4.3 by Chinese commercial cryptographic algorithm SM2/3/4, called Fabric-SM, and then evaluate the performance of Fabric v1.4.3 and Fabric-SM in various aspects, such as transaction throughput, execution time and transaction latency. The experimental results show that the performance of Fabric-SM is close to that of Fabric v1.4.3. Without any degradation in performance, Fabric-SM fully meets the requirements of Chinese government on the legality of cryptography. In a word, developers can develop various applications of Permissioned Blockchain based on Fabric-SM under the condition of safety and legality.

This paper focuses on the problem of object depth measurement in ultra-low altitude flight, which has always been a research hotspot in the field of machine vision. Compared with the previous object depth measurement method based on a single handcrafted feature point, the method based on feature line segments can effectively weaken the adverse effect caused by the feature matching error. However, since this method involves both image processing algorithms and the use of position data provided by navigation devices, various errors will be introduced. To this end, this paper points out the error sources of that may affect the accuracy of the object depth measurement results under this method. Then, through numerical simulation, the errors are simulated and their influence rules are analyzed, with the qualitative and quantitative results presented. Finally, we also give the allowable value range of each error under the minimum requirement that the object depth measurement error in ultralow altitude flight missions does not exceed 20%, which provides a reference and guidance for both algorithm selection and practical application.

The stability of a power electronic transformer (PET) under short-circuit faults is a crucial issue. In this paper, the output-terminal fault of the PET is considered, and a resistive-type superconducting fault current limiter (SFCL) is applied in the PET for guarding against the fault surge. The configuration structure of the PET integrating with the resistive-type SFCL is presented, and the SFCL's potential influence on the PET's fault current is theoretically probed. Using the MATLAB platform, the digital model of the PET-SFCL is built, and the three-phase fault scenario is simulated. The simulation results reveal that the resistive-type SFCL enables to very usefully refrain the three-phase fault currents of the PET. The fault current level can be stably suppressed lower than the safety margin, and the PET's stability during the fault is availably increased.

Aiming at the problem of airport surface behavior recognition in low illuminance environment, based on infrared monitoring image data, the eco (efficient fluctuation operators for tracking) target tracking method is proposed to track different targets on airport surface including non-cooperative targets. The method of extracting macro motion features from tracking results is studied. On this basis, the long-term motion features are learned recursively in the framework of LSTM (long short term memory) network to realize the recognition of typical human actions on the airport surface. The experimental results on the self-made video data set show that the model can make full use of the long-term motion information in the video sequence. It is suitable for the airport scene in low illumination environment, and has a certain recognition effect.

In recent years, with the popularity of Internet technology, criminals have used the Internet as a tool and platform for drug crimes, making online drug crime a new form of drug crime. The identification of network-related drug chat content is of great significance for preventing and combating cybercrime crimes. This paper collects 5,000 chat records and combines them with decision trees through neural networks to establish an efficient identification method to provide a reference for effective prevention and combat against cybercrime crimes.

Due to the importance of high precision depth data for driving assistance system, related methods to predict scene depth without ground truth have attracted more attention than ever. Therefore, this paper comes up with a novel unsupervised monocular depth estimation model. Based on the structure of Deep3D, the proposed model reduces the influence of the diverse scale of targets by processing input images as pyramid structure. Furthermore, the proposed model improves the ‘hole’ phenomenon in depth maps and solves it by up-sampling the intermediate disparity maps to the initial resolution. The experimental results show that this model achieves convincingly accuracy on classical KITTI dataset and excellent generalization capability on images randomly captured in real driving scene.

This paper proposes a visual tracking algorithm using reliability evaluation and feature selection mechanism in the framework of Correlation Filter (CF). Three additional modules are used to improve the current CF-based tracker. Firstly, a module of reliability evaluation is used to determine whether the current tracking result is reliable. Secondly, an updating module is used to determine whether to update the target model by comparing the reliability of current tracking result with historical average. Thirdly, a feature selection module is presented to select hand-crafted feature or deep convolutional feature according to the current tracking state. Experimental results on a benchmark dataset of fifty challenging test sequences show that the proposed method can reduce the interference of complex factors effectively.

Depth measurement technology plays an important role in the field of machine vision, and the depth measurement methods based on monocular vision have received more and more attention. However, previous depth measurement schemes based on single feature points and least squares calculation are susceptible to feature matching errors. To this end, this paper proposes a new object depth measurement method from monocular images based on feature segments. We use two images taken by the same camera and the pose information provided by GPS/IMU device to perform object depth measurement. The experiment based on the visual simulation software shows that the proposed method can improve the accuracy of the measurement results with a good robustness.

The recent popular tensor nuclear norm (TNN), a convex surrogate of tensor tubal rank, obtains promising results in tensor completion. Although the TNN-based model has shown its prominent capability of characterizing the global structure of tensors, it lacks the capability for preserving the abundant details of the target tensor. By integrating the global low-tubal-rankness and nonlocal self-similarity, we propose a novel tensor completion model, which recovers the global structural information by TNN regularizer while compensating for the details by plugging in a denoiser to express the nonlocal self-similarity prior. We design an alternating directional method of multipliers (ADMM)-based algorithm to solve the proposed model. Extensive experimental results on color images and fluorescence microscope images demonstrate the superiority of the proposed method over the compared ones.

As a kind of single-stage detection network, RetinaNet has excellent performance in object detection, but it still has some difficulties in small object detection in unmanned aerial vehicle (UAV) images. Based on RetinaNet, this paper uses improved ResNet50 as backbone to extract features. We increase the number of anchor ratios, but reduce the number of anchor scales so that the number of anchors won't increase too much. We add the shallower features to predict smaller targets. We also add Receptive Field Block after each predicted feature to increase the receptive field. We have changed the values of parameters in the loss function for better detection performance. The improved model effectively increases the detection rate of small targets so that it can be applied to pedestrian detection under UAV. In our UAV pedestrian dataset, the improved model increases average precision (AP) from 51.29% to 84.05%.

The drastic increase of the number of space debris in orbit around the earth poses a serious threat to operational spacecraft. Rapid small space debris detection with space-based surveillance platform is very important for spacecraft emergency avoidance at distance. Nevertheless, small space debris detection is a great challenge because of its fast movement and noise caused by cosmic rays in space-based surveillance platform. Inspired by strong pattern recognition capacity of deep learning, a deep convolutional neural network based small space debris saliency detection method for space-based surveillance system is proposed. First, the spatial contrast map of the space debris image is generated using local contrast method. Then, the spatiotemporal saliency information is captured incorporating with the above contrast map. The experimental results prove the applicability of the proposed method.

Visual tracking for aircraft is an important part of the airport surface surveillance. However, the current tracking algorithms do not perform well in complex environment like the airport. Aiming at this problem, this paper proposes a target tracking algorithm based on the correlation filter using deep conventional feature. Firstly, a convolution neural network is trained for the classification of aircraft. Then the shallow and deep features of the target are extracted by the network. Finally, these features are fused into the correlation filter tracking method. The proposed algorithm is compared with other trackers on ten video sequences with different weather conditions and different locations in the airport. Experimental results show that the proposed method can achieve high accuracy and success rate, and the overall performance is superior to other comparative algorithms.

Space debris is becoming a serious problem for spacecraft in norm operation. Visible sensor is mainly used to avoid potential collision from space debris in space-based optical surveillance projects. Nevertheless, it strongly relies on illumination of sunrise. Inspired by image fusion technology with deep learning, we propose an all-weather space debris recognition method with convolutional sparse representation to deal with different illumination. First, infrared and visible image which contained space debris are fused with convolutional sparse representation, then the space debris is recognized with deep convolutional neural network for fused image. The experimental results prove the applicability of the method.

In this paper, a novel flying object detection algorithm is proposed to assist with autonomous vision based sense and avoid system. Considering the specificities of flying objects detection, both the spatial and temporal context information are verified to be essential for ensuing the robustness of the algorithm. The algorithm is thus designed under a fusion architecture with the spatial and temporal context information. As for spatial context information extraction, the whole image is firstly sampled into a dense grid of image patches, then a pre-learned conditional random field (CRF) model is applied to generate the spatial probability map under a layered structure: CRF, sparse codes, bottom feature descriptors, and local image patches. As for temporal context information extraction, the motion cues are firstly detected by computing the forward back motion history image (FBMHI), then the foreground and background are further isolated by adaptive threshold selection. The probability map of the flying object presenting on the whole image is finally acquired by means of spatial and temporal context information fusion, and the flying object is detected on the basis of the spatial and temporal probability map. A number of aerial video sequences containing planes and drones are adopted for evaluating the effectiveness of the algorithm, and the results show the algorithm proposed in this paper performs favorable against other state-of-the-art techniques.

In this paper, we describe the basic strategy of Hex and analyze how to apply MCTS in Hex. Based on the parallel optimization of MCTS, an AI system for Hex is proposed. Experiments show that the system can make full use of prior knowledge to quickly reduce the scale of the search tree, and take advantage of computing resources through parallel programs to achieve an approximately linear increase in simulation efficiency. This system verifies the possibility of improving the efficiency of the search algorithm through the horizontal expansion of resources.

Hex game is a kind of board game with two players which has the rules that to seek the winner who firstly connect the two boundaries of the board with his own pieces. The decision of winning is the necessary factor of board game, and the effective decision algorithm is the premise to support the whole intelligence level of the game system. Based on the above inference, this article focuses on winning decision process in Hex, and especially for different winning situations to make different strategies. In order to meet the different winning situations, Union-Find Set and Deep-First-Search strategy will be used and hence get a valid algorithm, then, the complexity of algorithm will also be analyzed in the end.

Monte Carlo Tree Search (MCTS) is a heuristic search algorithm for some kinds of decision processes, it has many important applications to computer game domain. The most important thing in theory for MCTS is to reduce vari­ance of the estimate values of the tree nodes under the limited search scope due to the constraints of computing resource and time consumption. The original MCTS algorithm uses a random policy which usually does not converge. Some optimizations, like the UCT and the RAVE, can reduce variances partly through the use of episode information. In this paper, we introduce a new algorithm, which could produce low-variance estimate of node values via a statistical method called bootstrap. Experimental results show that the new algorithm significantly improves the theoretical performance of MCTS.

In this paper, we propose an online visual tracking algorithm for fused sequences via deep learning and adaptive Particle filter (PF). Our algorithm pretrains a simplified Convolution Neural Network (CNN) to obtain a generic target representation. The outputs from the hidden layers of the network help to form the tracking model for an online PF. During tracking, the moving information guides the distribution of particle samples. The tests illustrate competitive performance compared to the state-of-art tracking algorithms especially when the target or camera moves quickly.

This paper proposed a tracking algorithm based on sparse coding and spectral residual saliency under the framework of particle filtering. The proposed algorithm can be divided into three parts. Firstly, spectral residual is used to calculate a saliency map of the current frame and then compute the saliency score of each particle. Secondly, several particles are eliminated directly based on the differences between the saliency scores of the particles in the current frame and the target score in the prior frame. Thirdly, ScSPM is used to compute the observation vector for the rest particles and the tracking task is finished in the framework of particle filtering. Both quantitative and qualitative experimental results demonstrate that the proposed algorithm performs favorably against the nine state-of-the-art trackers on ten challenging test sequences.

Landlords is a Chinese card game for all ages, each game requires at least three players. An artificial intelligence service is growing rapidly to replace mankind to ensure that there are enough players online to play with people 24 hours a day. Since bidding is a vital part of this game, in this paper, we developed a bidding strategy by studying the real online gaming data from human's playing based on the Adaboost algorithm. We choose 12 kinds of card combinations as weak classifiers, and train them with Adaboost algorithm to get the final strong classifier. The strong classifier can make an evaluation for a given hand, which can be used for making decisions whether bid or not. Tested by different eigenvalue and card combinations data, we get an accuracy of 75% at most, basically reached the design requirements.

Evaluation function is one of the core components of the computer game system. The traditional way to build evaluation function is usually to use a number of features on the board extracted by human, according to experience and game rules, which gives a set of parameters, and design a static linear evaluation function to calculate the evaluation value of current situation. This paper proposes a method to dynamically adjust and optimize the parameters in the evaluation function by least mean squares, and implements a dynamic evaluation function. Experiments based on Surakarta chess show that the method of dynamic evaluation is effective.

As AlphaGo come into view, artificial intelligence for board games once again becomes the focus of the world, while the research for card games isn't deep enough. At present, the study of computer game algorithm of board games is more than in card games. One reason for this is the algorithms of card games are more difficult to implement than board games due to the imperfect information nature of the card games. We present a method that based on rules. We carry on research on card splitting, bidding, playing and following of Dou Dizhu. Experimental results show higher ratio of winning of our method than the method with simple strategies.

Blockchain is attracting more and more attentions from many kinds of fields, such as finance, industry, and theory researchers. Rather than viewing blockchain as a novel technology, we should think it as an innovation for managing digital society, which provides fundamental principles to support democratically distributed applications. Based on the design and application of the Energy Internet, this paper analyzes the core architecture of the initial blockchain technology, and combines the security of the public chain with the efficiency of the private chain, solving the poor efficiency problem of the initial blockchain by using the high efficiency of private chain. At the same time, it inherits the security and non-tampering of initial blockchain. This paper designs a new hybrid blockchain storaging mode that is Block Static Storage(BSS, internal structure is N+X hybrid blockchain, following can be referred to as N+X HBC),for purpose that improving the overall efficiency of the Internet running, achieving decentralized supervision, and provoding a credible, safe and efficient performance of the Energy Internet in the storage of its massive data, as well as a huge business system.

Automation of airport operations can greatly improve the ground movement efficiency. To this end, the taxi planning problem has been studied extensively in recent years, which aims to make full use of existing resources to reduce taxi time while ensuring safety. In this paper, we further study the aircraft ground movement speed profile design problem constrained by the surface time-based trajectory generated in taxi planning to facilitate precise guidance capabilities envisioned by the next generation air transportation system. A decomposed approach of two stages is presented to efficiently solve this problem. In the first stage, speeds are allocated at control points so that smooth speed profiles are expected to be found later. In the second stage, detailed speed profiles for each taxi interval are generated according to the allocated control point speeds which minimize the overall fuel consumption index. We present a swarm intelligence based solution approach for the first-stage problem and a discrete variable driven enumeration method for the second-stage problem. Experimental results show impressive performance of the presented methodology upon real world taxi speed profile design problems.

Intruder detection is a crucial problem in vision based Sense and Avoid (SAA) system. In this paper a deep feature learning based intruder detection algorithm is proposed. The intruder detection algorithm contains four parts: obtaining the test samples, creating the overcomplete dictionary, deep feature learning and determining the region of the intruder. The sliding window technique is adopted to obtain the test samples. The K-means Singular Value Decomposition (K-SVD) is used for overcomplete dictionary training. We employ the deep feature learning method on the basis of the dictionary for feature extraction. The support vector machine (SVM) is used to select the region of interest (ROI), and the region of the intruder is finally determined by merging the overlapping ROIs. The experiment results indicate that the algorithm is robust under different weather and illumination conditions and different angles of view.

In this paper, we address the problem of human pose estimation through a novel articulated Gaussian kernel correlation function which is applied to human pose tracking from a single depth sensor. We first derive a unified Gaussian kernel correlation that can generalize the previous Sum-of-Gaussians (SoG)-based methods for the similarity measure between a template and the observation. Furthermore, we develop an articulated Gaussian kernel correlation by embedding a tree-structured skeleton model, which enables us to estimate the full-body pose parameters. Also, the new kernel correlation framework can easily penalize undesired body intersection which is more natural than the clamping function in previous methods. Our algorithm is general, simple yet effective and can achieve real-time performance. The experimental results on a public depth dataset are promising and competitive when compared with state-of-the-art algorithms.

We propose a generalized Sum-of-Gaussians (G-SoG) model for statistical 3D shape modeling that is applied to human pose tracking from a single depth sensor. G-SoG generalizes the original SoG model by involving much fewer anisotropic Gaussians yet with better flexibility and adaptability. Both SoG and G-SoG are involved for pose tracking with different roles, where the former one is used to represent observed point cloud data through an efficient Octree partitioning, and the latter one is embedded with a quaternion-based articulated skeleton to create a standard human template model. We derive a differentiable similarity function between SoG and G-SoG that can be optimized analytically not only to learn a subject-specific articulated model but also to support sequential pose tracking where two additional terms (visibility and continuity) are also involved. Our algorithm is simple yet effective and can achieve real-time performance. The experimental results on a public depth dataset are promising and competitive when compared with state-of-the-art algorithms.

We propose a new point set registration method, Global-Local Topology Preservation (GLTP), which can cope with complex non-rigid transformations including highly articulated deformation. The registration is formulated as a Maximum Likelihood (ML) estimation problem with two topologically complementary constraints. The first is the previous Coherent Point Drift (CPD) that encodes a global topology constraint by moving one point set coherently to align with the second set. The second, which is inspired by the idea of Local Linear Embedding (LLE), is introduced to handle highly articulated non-rigid deformation while sustaining the local structure. Without any pre-segmentation, the newly introduced LLE constraint is particularly useful and effective when there are multiple non-coherent and nonrigid local deformations (i.e, the CPD assumption may be violated). We have derived the EM algorithm for the ML optimization constrained with both CPD and LLE terms, leading to the new GLTP algorithm. Experimental results on 2D and 3D examples show its accuracy and robustness in the presence of outliers and noise, especially in the case of highly-articulated non-rigid transformation.

Electromagnetic topology method is used to compute the inner electromagnetic field. In the first, it is established the topological interaction sequence diagram between external excitation source and internal electronic system through multiapertures. Then, the corresponding multistep iterative algorithm of apertures coupling is obtained. Finally, the CRIPTE software is used to calculate the electromagnetic effect from external field coupling to internal one and the terminal load of double conductor transmission lines. Numerical result illustrates that the interaction between apertures should be taken into consideration; more apertures be propitious to electromagnetic shielding under the same area; the farther distance between circuit and aperture, the better the shielding effectiveness; putting the electric circuit in the side of the cavity will achieve better shielding effectiveness than in the front of the aperture.

We present a multi-layer joint gait-pose manifold (multi-layer JGPM) for human motion modeling to enhance the representative capability of the original JGPM that represents gait kinematics by two variables. One is the pose to denote a series of stages in a walking cycle and the other is the gait to reflect the individual walking styles. Coupling pose and gait variables in the same latent space was shown effective for human motion estimation. However, the original JGPM is limited to one kind of human gaits, and its learning cannot be scaled up to a large dataset due to a high computational load. This work overcomes the limitations of the previous method by involving a multi-layer topology prior that is able to accommodate a variety of walking styles, leading to better motion synthesis results. Moreover, to learn multi-layer JGPM effectively and efficiently, we adopted two techniques, training data diversification and topology-aware local learning. The experimental results confirm the advantages and superiority of our proposed method over several existing Gaussian process-based motion models.

We present a new structure-guided joint gait pose manifold (JGPM) that represents gait kinematics by two variables. One is the pose to denote a series of stages in a walking cycle and the other is the gait to reflect the individual walking styles. Coupling pose and gait variables in the same latent space, such as a torus-like JGPM, was shown promising and effective for video-based motion estimation. However, the two-step learning used in torus-like JGPM is computationally expensive and it separates the optimization of pose and gait variables. This work overcomes the limitations of the previous method by developing a new structure-guided JGPM that is able to jointly optimize four variables in the same latent space, leading to a much compact parameter set while sustaining a comparable performance on video-based motion estimation, as well as a great potential for large-scale learning.

We evaluate recent Gaussian process (GP)-based manifold learning methods for human motion modeling, including our recently proposed joint gait and pose manifolds (JGPMs). Unlike most GP algorithms that involve either one latent variable or multiple independent variables in separate latent spaces, JGPMs define two variables jointly and explicitly in one latent space to represent a collection of gait data from different individuals. We develop a model validation technique to examine these GP-based algorithms in terms of their capability of motion interpolation, extrapolation, filtering, and recognition. Experimental results on both CMU Mocap and Brown HumanEva datasets show the superiority of JGPMs over existing GP algorithms for human motion modeling.

Deep carbonate of Tarim Basin has become an important study target of shoal reservoirs at present, and the reservoir capacity and its control factors of high-frequent meter scale cyclic carbonate strata have been the purposes of this study. This research has selected the upper layer of strata of Yingshan formation in Keping area of Tarim Basin as the target stata, and analyzed by anatomying its outcrop, sampling and studying microscopic flakes. The result shows that the strata of Yingshan formation has deposited as high-frequent meter-scale cyclic sequence, and been an aggradational pattern between parasequence sets. Each parasequence set contains four parasequences which limestone was thickening during depositing period, the external style and internal composition between parasequence sets and parasequences are apparent self similarity. Calcarenite and bioclastic debris are the main particle types of shoal reservoirs, and the size gradually has increased up. Cement and means of contact between particles are consistent with the aggradational high-frequency sequences. All of the above fully have proved that the depositional energy increases from the parasequence sets A to C, and the high-energy deposition environment provides the basic condition for forming excellent shoal reservoir. Cracks and fissures are the main reservoir spaces of Yingshan formation shoal reservoir, but deep diagenesis shortly has not conducted to form effective dissolution intergranular pore, and reduced the capacity of quality shoal reservoir as the prerequisite for high-energy environment. As a result, the tectonic uplift has been the leading factor to control Yingshan formation to be II category reservoir.

Although data parallelism is a well-known computational model, there are few programming systems that are both easy to program (for simple applications) and able to work across administrative domains. For data sets (e.g., collections of image data) that are often inherently distributed, there is a need for a simple data-parallel programming system. We describe the design, implementation, and an evaluation of Trellis-SDP, a simple data-parallel programming system that facilitates the rapid development of data- intensive applications. Trellis-SDP is layered on top of the Trellis infrastructure, a software system for creating overlay metacomputers: user-level aggregations of computer systems. Trellis-SDP is based on file-level data parallelism and provides a Master-Worker programming framework in which the worker components can run self-contained, new or existing binary applications. We evaluate our programming system with a non-trivial seismic data processing application.