Search

Robust array beamforming is a challenging task in radar, sonar and communications due to the influence of direction of arrival (DOA) mismatch and sensor position errors. However, how to enhance the robustness of beamforming is a key issue in antenna arrays. The current paper focuses on a novel approach called the improved chicken swarm optimization (ICSO) method to settle the optimization model of conventional linearly constrained minimum variance (LCMV) based on support vector machine (SVM) to against the mismatch problems as well as control the sidelobe level (SLL). As far as the ICSO method is concerned, considering that the particle swarm optimization (PSO) algorithm has outstanding convergence performance in the early iteration, the dominance of the alpha wolf in the grey wolf optimization (GWO) algorithm and the innovative mutual attraction mechanism in the firefly algorithm (FA), and we introduce these three strategies into the solution update method of conventional chicken swarm optimization (CSO) algorithm for achieving better optimization capability. Moreover, an operation of removing duplicate solutions is proposed to enhance the utilization of the population. In terms of the SVM-based LCMV beamforming algorithm, we adopt the so-called linear $\varepsilon $ - insensitive loss function to reconstruct the final cost function of LCMV by penalizing the errors between the actual and ideal array responses. Finally, we conduct simulations to evaluate the performance of the swarm intelligent optimization algorithms under an ideal scenario without mismatch and an actual scenario with the mismatch, respectively. And the results demonstrate that the developed ICSO algorithm obtains excellent robustness for different scenarios compared to PSO, FA, GWO and CSO optimization algorithms.

Relation extraction is a popular topic in natural language processing. This survey will introduce several major relation extraction methods, from feature and kernel-based to neural networks. We will discuss the basic idea and the performance of these methods and hope that this article will give readers a general concept of relation extraction.

In this research, we aim at supplementing named entities, such as food, omitted in the procedural text of recipe data. It helps users understand the recipe and is also necessary for the machine to understand the recipe data automatically. The contribution of this research is as follows. (1) We construct a dataset of Chinese recipes consisting of 12,548 recipes. To detect sentences in which food entities are omitted, we label named entities such as food, tool, and cooking actions in the procedural text by using the automatic recipe named entity recognition method. (2) We propose a method of recognizing food from the attached images. A procedural text of recipe data is often associated with an image, and the attached image often contains the food even when it is omitted in the procedural text. Tool entities in images in recipe data can be identified with high accuracy by conventional general object recognition techniques. On the other hand, the general object recognition methods in the literature, which assume that the properties of an object are constant, perform not well for food in recipe image data because food states change during cooking procedures. To solve this problem, we propose a method of obtaining food entity candidates from other steps that are similar to the target step, both in sentence similarity and image feature similarity. Among all the 246,195 procedural steps in our dataset, there are 16,593 steps in which the food entity is omitted in the procedural text. Our method is applied to supplement the food entities in these steps and achieves the accuracy of 67.55%.

In order to effectively settle the issue of steering vector mismatch caused by non-ideal factors such as array element position disturbance and direction of arrival (DOA) estimation error, an improved chicken swarm optimization (ICSO) method is proposed to enhance the robustness of the beamforming algorithm. This method utilizes the characteristic of support vector machine (SVM) structural risk minimization, and combines it with the conventional diagonal loading (DL) beamforming algorithm. By selecting the appropriate loss function and penalty parameter, the ICSO method is finally employed to solve the standard support vector regression beam optimization model. As for the ICSO method, due to the particle swarm optimization (PSO) algorithm with fast convergence speed in the early stage of iteration, the firefly algorithm (FA) with novel attraction mechanism and the grey wolf optimization (GWO) algorithm with similar internal hierarchical order to the chicken swarm, these 3 algorithms are fully integrated into the conventional CSO algorithm. Moreover, the operation mechanism of removing duplicate solutions is introduced to increase the diversity of the population, thereby improving the optimization performance of the algorithm. The simulation results demonstrate that when the perturbation deviation of the array element position is 0.02λ, 0.20λ, and the DOA estimation error is 2°, the proposed method has better beam pointing and lower sidelobe level.

Recent years have witnessed great progress of object detection. However, due to the domain shift problem, applying the knowledge of an object detector learned from one specific domain to another one often suffers severe performance degradation. Most existing methods adopt feature alignment either on the backbone network or instance classifier to increase the transferability of object detector. Differently, we propose to perform feature alignment in the RPN stage such that the foreground and background RPN proposals in target domain can be effectively distinguished. Specifically, we first construct one set of learnable RPN prototpyes, and then enforce the RPN features to align with the prototypes for both source and target domains. It essentially cooperates the learning of RPN prototypes and features to align the source and target RPN features. Particularly, we propose a simple yet effective method suitable for RPN feature alignment to generate high-quality pseudo label of proposals in target domain, i.e., using the filtered detection results with IoU. Furthermore, we adopt Grad CAM to find the discriminative region within a foreground proposal and use it to increase the discriminability of RPN features for alignment. We conduct extensive experiments on multiple cross-domain detection scenarios, and the results show the effectiveness of our proposed method against previous state-of-the-art methods.

The hesitant fuzzy set (HFS) is widely applied in actual multi-attribute group decision-making (MAGDM) problems. It can depict experts' hesitant evaluation information with the membership degree consisting of several possible values. Most existing methods based on HFSs only focus on the final integrated information by different kinds of aggregation operators but fail to provide detailed comparisons between alternatives. They are essentially result-oriented static decision-making methods, based on which, the decision-making results may be inconsistent with reality. However, there is no process-oriented research on hesitant fuzzy information. The decision field theory (DFT) is a dynamic decision-making method and can better simulate the uncertain decision-making process. Thus, this paper integrates the HFS into the DFT and proposes a new decision-making method named as hesitant fuzzy decision field theory (HFDFT) to fill this vacancy. First, we define the hesitant fuzzy momentary preference function and other parameters in HFDFT. After that, for the MAGDM problems with incompletely known attribute weight information, the programming model is used to determine the weights of attributes. Then, the group decision-making method based on HFDFT is presented. Moreover, we apply the proposed HFDFT method to a case about route selection of the Arctic Northwest Passage. Two traditional methods based on the score function and the correlation coefficient, respectively, are further implemented for comparisons to illustrate the validity of the proposed HFDFT method.

A data processing method for correcting the frequency sweeping nonlinearity of a tunable laser in an optical frequency domain reflectometry system is presented. The proposed method calibrates the positions of all the data points by the time-scale factor determined by the zero-crossing points of the auxiliary interferometer. Then, the signal of the main interferometer is interpolated with the calibrated results. In experiments, a spatial resolution of 0.17 mm over a measurable range of 155 m has been obtained even when the optical path difference of the auxiliary interferometer is shorter than that of the main interferometer. This method could resolve the problem that measurable range is limited by the length of delay fiber in the auxiliary interferometer and significantly decrease the requirement of sampling rate.

Smart contracts are programs that automatically execute on the blockchain system such as Ethereum. Everybody can write and deploy smart contracts on Ethereum, which causes a large collection of similar contracts via code reuse. In practice, code reuse in smart contract may amplify severe threats like security attacks, resource waste, etc. In this paper, we conduct an empirical study of code reuse in smart contracts for understanding the code reuse practice in the smart contract ecosystem. We first collect 146,452 open-source smart contract projects from Ethereum and then perform a detailed analysis. We first study how often the smart contract projects reuse and then we identify the top reused smart contracts and analyze how the developers revise smart contracts during reuse. Our research suggests that the code reuse in smart contract is quite frequent because about 26% contract code blocks are reused and the average time of reuse is 14.6. And the top reused contracts are almost all related to ERC20 token, which reveals that the current smart contract ecosystem is relatively homogenous. At last, we summarize 7 common types of code revision in smart contracts.

Performing autonomous exploration and exploitation is essential for un- manned aerial vehicles (UAVs) operating in unknown environments. Often, such missions involve first building a map of the environment via pure exploration and subsequently exploiting it for specific downstream tasks. But, conducting separate exploration and exploitation steps is not always feasible in practice. In this paper, we develop a novel exploration approach enabling exploration and exploitation in a single step for an area-of-interest (AoI) search task. The basic idea is to employ a probabilistic information gain map, called a belief map, as a prior to guide the exploration trajectory, while efficiently reducing false positive information in the process. The approach is composed of three layers. The first is an information potential layer to decide the exploration direction for the UAV. Next, the proximity layer exploits detected AoI by exploring their proximal areas. The last layer, a forced movement layer, is responsible for enabling the UAV to escape local maxima caused by the previous layers. We tested the performance of our approach in two different tasks relative to two exploration methods published in the literature. The results demonstrate that our proposed approach is capable of navigating through randomly generated environments and covering more AoI in fewer time steps compared to the baselines.

In CrowdIntell Network, there are various transactions between different Digital-selfs, and these transactions are affected by the mind of the intelligence subjects. Mining and analyzing the mind of the digital-selfs is conducive to the better decision-making of the Intelligent subjects, and promotes the conclusion of transactions. We focused on how to model the interests of Intelligent subject in a complex environment, and how to use interests to make the recommendation. Considering the relevance and sequential characteristics of the behavior of the Intelligent subject in CrowdIntell Network is of great value for predicting future behavior. Most of the previous work was directly modeling the sequence patterns on interactive history, and then training a DNN model to learn the interest representation. Recent research focuses on interest mining by embedding both sequential and original characteristics of behaviors. However, in a real environment, the interests of users (for the convenience of description, we simplify intelligent subjects to users) are complex and diverse (e.g., a certain user likes documentaries and also likes science fiction movies), and a user may has multiple interests. In order to solve this problem, a Multi-interest Aware Recommendation in CrowdIntell Network is proposed, which involves the embedding layer, two-stage feature extraction layer and full connection layer. Specifically, the model is firstly constructed and trained to learn the user's multi-interest representation by considering the interaction between items in the behavior sequences, and then the convolutional attention mechanism is used to extract local features on the interests “image” as a higher-order user interest representation. Experiments on public datasets show that our method outperforms state-of-the-art sequential recommendation methods.

Hardware design, especially the very large scale integration(VLSI) and systems on chip design(SOC), utilizes many codes from third-party intellectual property (IP) providers and former designers. Hardware Trojans (HTs) are easily inserted in this process. Recently researchers have proposed many HTs detection techniques targeting the design codes. State-of-art detections are based on the testability including Controllability and Observability, which are effective to all HTs from TrustHub, and advanced HTs like DeTrust. Meanwhile, testability based detections have advantages in the timing complexity and can be easily integrated into recently industrial verification. Undoubtedly, the adversaries will upgrade their designs accordingly to evade these detection techniques. Designing a variety of complex trojans is a significant way to perfect the existing detection, therefore, we present a novel design of HTs to defeat the testability based detection methods, namely DeTest. Our approach is simple and straight forward, yet it proves to be effective at adding some logic. Without changing HTs malicious function, DeTest decreases controllability and observability values to about 10% of the original, which invalidates distinguishers like clustering and support vector machines (SVM). As shown in our practical attack results, adversaries can easily use DeTest to upgrade their HTs to evade testability based detections. Combined with advanced HTs design techniques like DeTrust, DeTest can evade previous detecions, like UCI, VeriTrust and FANCI. We further discuss how to extend existing solutions to reduce the threat posed by DeTest.

Wilderness Search and Rescue (WiSAR) requires navigating large regions - often in rugged, remote areas - searching for missing people or animals. Because of the large regions and potentially limited mobility of ground vehicles, WiSAR missions are frequently carried out with the help of Unmanned Aerial Vehicles (UAVs). However, the ability to autonomously execute WiSAR remains an unsolved challenge. In this paper, we take advantage of Deep Reinforcement Learning (DRL) to develop an autonomous WiSAR controller for UAVs. We improve the learning and understanding of a UAV agent to explore a partially observable environment in search of a victim trapped in the wild. The proposed approach breaks up this difficult problem into 4 sub-tasks: tractable mapping of the environment in small regions, region selection, target search, and region exploration. Quad-Tree is utilized offline to decompose the environment map into smaller, tractable maps. Then, an efficient cost function is repeatedly computed to determine the best target region to search in each iteration of the process. Recurrent-DDQN and A2C algorithms are trained to generate optimal policies for the target search and regions exploration tasks, respectively. We tested our approach against a baseline of a hard-coded policy of navigating the map in a zigzag fashion and another baseline of using the same sub-tasks but instead of using the DRL algorithms, randomly selecting an action at each time step. The results demonstrate that our proposed approach is capable of navigating through 25 randomly generated environments and finding the missing victim faster than the baselines by 46%.

With the transformation of Internet technology from traditional PC terminal to personal mobile terminal, Chinese e-commerce industry has achieved unprecedented development. Live streaming technology not only provide more abundant marketing mode for merchants, but also creates a more convenient environment for its audiences. Exploring the factors that affect users’ purchasing behavior in live broadcast scene can provide practical guidance for the next optimization of live-stream technology. Meanwhile, it is also important for the further development of e-commerce industry in multimedia environment. From the perspective of users, this paper explored and analyzed the application of computer multimedia technology in e-commerce, and draws the following conclusions: users’ age is negatively related to their willing to purchase; while, users’ network application skill, live-stream platforms’ reputation, ease of use and price discount of the recommended products are all positively related to users’ purchasing intention. To sum up, we need to build a more convenient live-stream environment with the help of the multimedia computer technology, and then further reduce the "technical threshold" of the elderly users.

Edge computing can provide quality assurance for multimedia services. In order to protect the environment, many communities have begun to use green energy, so we add this status to the formulation of the computation offloading mechanism. In this paper, we investigate the computation offloading mechanism for multimedia services in green smart community. First, we introduce a dynamic adjustment mechanism for unit cost of local energy consumption, which makes our algorithm more flexible. Then, we propose a computation offloading mechanism, which aims to decouple the optimization problem and minimize the system cost. We model the computational task offloading decision problem as a distributed potential game, and solve the resource allocation problem using mathematical methods and theories such as Lagrange multiplier method, KKT condition, gradient descent, bisection method, and we also design a CBPA (Computation resource and bandwidth and transmission power allocation) algorithm to optimize edge computing. Finally, simulation results show that the computation offloading mechanism proposed in this paper is effective.

A cross-border supply chain network is a complex network system that involves many participants, spans a large geographical space, with a high degree of complexity and uncertainty. Based on the theory of supply chain management, network topology, and system engineering, this paper conducts a detailed analysis from the connotation and characteristics of cross-border supply chain network, topological structure model and logistics network structure model, reliability and influencing factors, and describes the basic characteristics of cross-border supply chain network.

Frozen waves have the good characteristic for resisting the diffraction and absorption in long distances absorbing medium, on the other hand, vortex carrier can improve the information carrying capacity of communication system. So, vortex frozen waves can be one of the effective waves to solve the low transmittance of light waves in seawater. However, when such light waves as information carrier of underwater optical communication system travel through turbulent waters, they are inevitably disturbed by turbulence and thus reduce their ability to resist absorption and diffraction. Therefore, the absorption and diffraction resistance of vortex frozen waves in turbulent waters is a topic worthy of discussion. Thus by the turbulence spectrum of seawater and the theory of frozen waves, we establish a transmittance model for finite energy vortex frozen waves. By this model, we analyze the diffracting- and absorption-resistance of vortex frozen waves from the beam transmission wavelength, beam waist, beam orbital angular momentum quantum number, inner scale of turbulence, the "temperature structure" constant and the ratio of temperature and salinity. Through the above analysis, we get the optimal beam parameters under certain turbulence conditions.

Retroflex and non-retroflex consonants in Mandarin (z[ts], c[ts'], s[s], zh[ts]; ch [ts'], sh[s]) are difficult to produce for many Chinese as a second language (CSL) learners with different mother tongues. Aiming at developing an efficient way to help them to master the sounds, this study adopts the idea of Oral Motor Exercise (OME) to devise a speech production training method. The method consists of two steps: the first is non-speech OME, in which muscle groups of jaw, lips and tongue, essential to the production of target consonants, are intentionally exercised; the second is speech OME, in which monosyllables with target consonants are practiced. 30 participants took part in the experiment. The results showed that both non-speech and speech OMEs could effectively reduce subjects' consonant errors in a short time. When the two steps were accomplished, the trainee achieved the best performances. These indicated the effectiveness of the proposal.

Frequency modulated continuous wave synthetic aperture radar (FMCW SAR) has become one of the main SAR research directions for its small size, light weight and high resolution, which is suitable for air-borne or UAV-borne SAR imaging. In this paper, a FMCW SAR structure is proposed, the transmitter and receiver are placed at the same flying path, separated with a constant distance to attain good isolation. The simulation result is presented for SAR imaging.

Deep features extracted by convolutional neural networks have demonstrated state-of-the-art performance on image classification and retrieval tasks. Firstly, this work presents a simple and efficient method to select the region of interest based on semantics of features. Then, this work spatially weight deep convolutional features based on analyzing the relations between the self-aggregation vector and the indicator matrix. Based on the two strategies mentioned above, we product a compact global image representation of all images and show the proposed approach outperforms previous strategies based on pre-trained networks on a couple of benchmark datasets, i.e., Oxford dataset and Paris dataset.

Due to the buried structure of the cross subway transfer station, effect of piston wind in the station will influence fire smoke spread in underground platform. In this paper, large eddy simulation (LES) method is used to study the fire simulation in a cross-typed metro transfer station. Smoke diffusion, temperature and CO concentration distributions, wind speed at the stairway are predicted for different smoke control modes. Based on the on-site measurement results of piston wind at platform in a condition with non-stopping train, the influence of deep buried structure and piston wind effect on fire smoke spreading rule of underground platform is studied. Results show that the buried structure of transfer station affects the natural air supplement of the ground floor platform. The piston wind effect has an adverse effect on the air distribution in the fire. In addition, air supplement volume at stairway cannot meet the requirements of the standard of metro design.

A distributed vibration-sensing system based on phase-sensitive optical time-domain reflectometry (Φ-OTDR) and distributed feedback interferometer (DFI) with broadband frequency response and high spatial resolution has been proposed and demonstrated. A couple of narrow line-width lasers with different wavelengths were used to form the light source, which were consistent with two channels of dense wavelength division multiplexer. Both functions of Φ-OTDR and DFI could work synchronously without influence between each other. The characteristic of broadband frequency disturbance event could be fully captured and identified by the DFI, while the corresponding location could be determined by the Φ-OTDR scheme. Unlike ordinary interferometer, the proposed DFI scheme could make single-end monitoring without the installation of extra accessories. Furthermore, broadband frequency response range and precise locating of the disturbance event could be achieved at the same time. Experimental results have shown that spatial resolution of 10 m and frequency response up to 1 MHz have been obtained over 2.16-km-sensing fiber, which have proved the validity of the proposed method.

The two-dimensional classical multiple signal classification (MUSIC) algorithm is commonly used in the direction of arrival (DOA) estimation. However, it has a large amount of computation and slow computation when the spectral peaks are being searched. In view of these problems, an improved chicken swarm optimization (ICSO) algorithm with adaptive capability is proposed. In this scheme, the initial population is constructed by the theory of good point set, and the inertia weight is dynamically adjusted by introducing the feeding speed factor and aggregation degree factor. Finally, the inertia weight is added to the position updating formula of the hen. The experimental results indicate that the new method not only ensures the searching success rate, but also reduces the complexity of the computation. What's more, the system has a quicker convergence and a better stability.