Search

The scattering mechanism interpretation of oriented manmade targets especially those with large orientation angles is a challenging task. Recently, to fit the cross-polarization and off-diagonal terms, physically meaningful double-bounce and odd-bounce scattering models have been developed by modeling their independent orientation angles. Nevertheless, parameters inversion of these models is a nonlinear optimization procedure. The main purpose of this work is to convert the nonlinear model inversion procedure to a linear solution while inheriting the generalized models. By disclosing the latent relationship between the polarization orientation angle and double-bounce orientation angle, a double-bounce orientation angle inversion method is developed. Then a refined double-bounce scattering model is proposed and a fast general model-based decomposition is established. The comparison experiments are performed on PiSAR data. Compared with the state of art approaches, the proposed decomposition method achieved improved decomposition performance from both visual and quantitative investigation for oriented built-up areas.

Robust synchronization of chaotic systems becomes a hot topic in scientific and engineering fields because of the ubiquitous existence of time-variant external disturbances in complex application scenarios. In contrast with existing studies that the resultant synchronization error has a supremum or even diverges under the influence of time-variant external disturbances, this paper proposes a disturbance rejection zeroing neurodynamic (DRZN) approach and its related controller for the robust synchronization of chaotic systems against time-variant external disturbances. The controller designed by the proposed DRZN approach distinctively features the rejection of external disturbances with the generated synchronization error being convergence toward zero. Theoretical analyses guarantee that the DRZN approach and its related controller inherently possess robustness. Moreover, numerical studies including three examples substantiate the effectiveness of the proposed DRZN approach and its related controller for the synchronization of chaotic systems against the time-variant external disturbances. Comparisons with existing approaches, e.g., the conventional zeroing neurodynamic (CZN) approach and the linear-active control (LAC) approach, show the superiority of the proposed DRZN approach. Extensive tests further verify that the proposed DRZN approach possesses the outstanding anti-disturbance performance, and thus is suitable for the practical applications with time-variant external disturbances.

Recent years have seen a rapid development of film and television industry, Concurrent with this increase has been a steady rise in user demands on the quality of video content on Internet. This paper mainly introduces HanLP natural language processing package and TextRank algorithm, explains its operation principle in detail by comparing PageRank and TextRank, so as to further study the automatic abstract generation technology of entertainment film and television news. Based on the natural language processing toolkit HanLP, this paper improves and strengthens the abstract generation technique and realizes the core sentence generation. We conduct comprehensive experimental studies. The experimental results show that this method used in this study is of higher quality and more readable than state-of-the-art methods. This technology can also effectively analyze important sentences in film and television news, generate readable abstracts, provide convenience for readers to extract the core fragments of the text, and has certain application value.

Accurately prediction of building load is essential for energy saving and environmental protection. Exploring the influence of building charac teristics on heating and cooling load can improve energy performance from the design phase of the construction. In this paper, a prediction model of construction heating and cooling loads is proposed, which based on improved Particle Swarm Optimization (IPSO) algorithm and Long Short-Term Memory (LSTM) neural network model. Firstly, the characteristic variables are extracted and evaluated by Spearman’s correlation coefficient method; Then the prediction model based on the LSTM neural network is constructed to forecast building heating and cooling load. The IPSO algorithm is adopted to deal with the trouble that manual work cannot precisely adjust parameters. In this method, the optimization capability of the PSO algorithm is improved by changing the updating rule of inertia weight and learning factors. Finally, the parameters of the LSTM neural network are taken as IPSO optimization object to realize the combination of the two algorithms and improve the prediction accuracy. In the experimental stage of this paper, a variety of algorithm models are compared, and the results show that IPSO-LSTM can get the best results in the prediction of heating and cooling load.

The ability of expression includes communication ability, generalization ability, leadership ability, communication ability, etc. As the basic ability of talents, it is in a very important position in the talent evaluation system. However, the current evaluation of expression ability is still at the primary level of the interviewer’s decision, which requires a lot of manpower and time costs, and cannot be quantified, and can only be divided into levels by experience. Aiming at this problem, the expression ability evaluation project proposed a new expressive ability evaluation program called Problem Kit System. By participating in the evaluation, users can quantify their expression ability level. Experiment results show that the system can quantify the expression ability of the evaluator to a certain extent, and the number of samples whose scores are within 5 points of the expert’s rating exceeds 80%.

Generalised spatial modulation (GSM) activates more than one transmit antenna to send information each time and can achieve higher transmission rates than conventional spatial modulation (SM). Because the GSM information matrix cannot have a unitary matrix structure, the usual differential scheme cannot be used directly in GSM systems. A new scheme based on generalized spatial modulation is proposed(NGSM), in which each transmit frame consists of a reference block and normal blocks. Every normal block is differentially coded based on reference. Reference blocks are allocated more power than normal blocks. The performance of the NGSM system is analysed, and an analytical expression for a tight bound of the average bit error probability (ABEP) is derived. A sub-optimal low-complexity detection algorithm, message passing (MP), is presented to reduce the computational complexity of detection. Simulation and theoretical results demonstrate that the proposed NGSM scheme has only 0.5 dB of ABEP performance loss as compared to coherent GSM when the frame length is 400. The MP algorithm achieves near-maximum likelihood detection accuracy, while reducing complexity by 70%.

The female reproductive potential in mammal depends on the numbers of primordial follicles that are formed before and after birth. The formed follicles are subsequently consumed through repeated ovulation cycles. The classical primordial follicle counting is a very complicated, tedious, and time-consuming task that relies on expert identification. Moreover, the follicles in the neonatal ovary are composed of primordial follicles and a few primary follicles. Their large number and small size undoubtedly increase the difficulty of identification and counting the number of follicles. To effectively solve the problem of the number of follicles, we used deeplabv3+ as the neural network structure for follicle segmentation. On this basis, the watershed algorithm was used to eliminate the boundary problem between follicles and follicles. Finally, we counted the connected areas in the images. In this poster, we aim at addressing the problem of counting ovarian follicles in biomedical images, achieving the cross-application of biomedical images processing and artificial intelligence.

This research aimed to examine the relationships between college education outcome expectation, proactive personality, career adaptability and learning motivation. 576 Chinese college students participated in this study. The results were as follows: (1)Significant differences between college students with and without siblings were found in the dimension of career adaptability's total score, and also in concern, control, curiosity and confidence dimension of career adaptability, as well as in the total score of learning motivation and its dimension of intrinsic motivation; whereas no significant difference existed in college education outcome expectation and proactive personality between students with and without siblings; (2)Structural equation modeling revealed that both college education outcome expectation and proactive personality significantly predicted learning motivation; (3)Career adaptability partially mediated the influence of college education outcome expectation and proactive personality on learning motivation. According to these findings, suggestions for college education, especially on career development education and learning motivation stimulating, were proposed and discussed.

In fingerprint positioning, the construction of a fingerprint database has a crucial impact on positioning. Because of the wide outdoor environment, the accurate construction of a traditional fingerprint database requires technologies such as large-scale drive testing, which consumes a lot of manpower and material resources. In this paper, Universal Kriging interpolation method is proposed to quickly construct outdoor fingerprint database for solving the problem of time-consuming and laborious. In order to detect the performance of the algorithm, the Reference Signal Receiving Power (RSRP) at the sampling point is calculated by using the Standard Propagation Model (SPM). The experiment is done using Universal Kriging algorithm and Inverse Distance Weighted algorithm to interpolate the database with 30% sampling points. The results show that the fingerprint database constructed based on the Universal Kriging interpolation algorithm is higher accuracy.

Although twin support vector machine (TWSVM) is widely used in practice, it only takes the empirical risk minimization principle into consideration while neglects the data distribution information, which seriously limits the classification performance. In view of this, robust minimum class variance twin support vector machine (RMCV-TWSVM) is proposed. Compared with TWSVM, RMCV-TWSVM has better generalization ability because it takes both the empirical risk minimization principle and data distribution characteristics into consideration. While TWSVM and RMCV-TWSVM always neglect the local characteristics of each class sample. Based on the above analysis, this paper presents least squares twin support vector machine based on manifold-based within-class scatter (MWCS-LSTSVM), which not only focuses on the empirical risk minimization principle and data distribution characteristics, but also preserves the local manifold structure of each class sample. At the same time, the solutions of MWCS-LSTSVM comes directly from solving a pair of systems of linear equations rather than solving two quadratic programming problems. Thus, MWCS-LSTSVM has low time complexity. The experimental results show the significant advantages of proposed approach in both classification accuracy and computational time.

Twin support vector regression (TSVR) is a regression algorithm based on the support vector regression (SVR) and the spirit of the support vector machine (TWSVM) . However, some feature selection algorithms of support vector regression, such as recursive feature elimination, can’t be applied to TSVR, so a recursive feature selection method based on TSVR is proposed. By analyzing the weights, the e -insensitive upper and lower bound functions in TSVR are analyzed. The two weight vectors are merged, and the weight vector is sorted and deleted with reference to the recursive feature elimination (RFE). The experimental results on several UCI datasets demonstrate demonstrate the effectiveness of the algorithm on feature selection and improves the regression performance.

Polarimetric target decomposition is a powerful tool for polarimetric synthetic aperture radar (PolSAR) data interpretation and utilization. Although many advances have been reported, scattering mechanism ambiguity still occurs for manmade targets with large orientations. Recently, a general model-based decomposition has been established to mitigate this ambiguity by modeling the orientations of manmade structures. However, the optimal solutions are obtained with nonlinear optimization which usually costs a large computation time. This work aims at maintaining the generalized scattering models while significantly reducing the computation burden. The main contributions are two-fold. Firstly, the generalized scattering models are adapted to fit the actual observed situation. The second contribution is the parameter inversion. In proposed decomposition procedure, the volume scattering power is derived through an eigen decomposition problem. Then, the helix scattering power is obtained through the imaginary part of a coherency matrix element. The next two scattering components of odd-bounce and double-bounce are calculated through a set of linear equations. As a result of extra freedoms, the roll-invariant feature, Alpha angle from H/alpha/A decomposition is selected as a branch condition to discriminate between dominant odd-bounce and double bounce scattering mechanisms. The comparison experiments are carried out on airborne PolSAR data sets. The experiment results demonstrate that the proposed method achieved improved decomposition performance and greatly reduced computation time.

Blockchain is an emerging decentralized architecture and distributed computing paradigm underlying a large number of cryptocurrencies and has recently attracted many high-tech institutions. It has a wide range of applications in various fields. With the continuous development of higher education, the curriculum evaluation system plays a vital role in the construction of college curriculum. Owing to the defects of non-transparent and centralization in college curriculum, a new frame based on blockchain is proposed in this paper. The proposed framework separates the curriculum evaluation from the school educational system and uses the characteristics of the blockchain to achieve the validity and openness of the evaluation results. Performance analysis and practical instances show the proposed frame can improve the authenticity and effectiveness of the evaluation.

Ship-crossing is a common behavior within inland waterways, which will influence the navigation safety for both own crossing-ship and ships using channels. This paper proposes a human-like decision-making approach based on Bayesian belief network (BBN) for ship-crossing by considering ship behaviors and maneuverability. The input variables consider the motion information of ships and the predicted information in future moment, while the evaluation variables are used for determining the intention of crossing-ship. The proposed model is validated in the real crossing scenario in the midstream of Yangtze River. From the results of scenario of ship-crossing, the proposed approach can be used for navigation assistance of ship crossing in busy waterways.

The crossing of a channel is a typical scenario for the ferries in the Yangtze River, which increases the navigation risk for both the ferry and the ships navigating in the channel. From the perspective of crossing behavior, this paper proposes a predictive method based on Hidden Markov Model (HMM) to analyze the dynamic process of crossing. The observation factors are established by analyzing expertise of captains of passenger ships and ferries. The state transfer matrix between safe crossing and unsafe crossing, together with the emission matrix between hidden states and observation, is proposed by using statistical data. The proposed method is applied to the real crossing scenario of a WUDU2HAO ferry in the Changzhou port of Yangtze River as a test. The result state that the proposed method for rationality prediction of crossing behaviors of ferries is reasonable.

Roll-invariant polarimetric features which are independent of target orientation angle along the radar line of sight are popularly adopted in many radar application fields. During the development history of radar polarimetry, a number of roll-invariant polarimetric features have been reported in terms of different polarimetric matrix formulations, different target decomposition approaches and different scattering mechanism interpretation tools. Currently, there is lack of a comprehensive summary of these valuable roll-invariant polarimetric features. Also, their inner-relationships need to be further disclosed. Finally, deep investigations of their application potentials are essentially necessary. This work is dedicated to these aforementioned issues and a survey of roll-invariant polarimetric features is carried out.

Course control is one of the key research points for intelligent navigation of ships. Based on the optimization characteristics of Beetle Swarm Optimization (BSO) algorithm, this paper designs a ship course online self-optimization controller. Firstly, the ship MMG model is established, and the rudder turning rate constraint is introduced to establish the course control objective function. Then, based on the PSO algorithm, the BAS algorithm is improved to the BSO algorithm, and the BSO-PID algorithm is used for online course control. Through a series of simulation experiments, it shows that the method of self-optimizing PID parameters by BSO algorithm has better control effect on ship course control. Furthermore, it lays the foundation for further research in combination with other algorithms.

Conventional marine hydraulic motors may not be suitable due to the effects of polar temperature environments. In this paper, for the same type of marine hydraulic vane motor, the friction and wear tests were carried out on the vanes of different materials to analyze the influence of low temperature environment on the friction and wear properties of different materials. The material of conventional marine hydraulic motors vanes was spring steel and the material of polar marine hydraulic motors vanes was martensitic stainless steel. The sliding wear tests of the two materials were performed against ductile iron discs by using wear tester under different temperature conditions respectively. In order to analyze the wear mechanism of the friction test specimen under different conditions, the surface morphology of the friction test was carried out by Scanning electron microscope (SEM). The results showed that the coefficient of friction between the pin samples of the two materials and the disk samples of ductile iron didn’t change linearly as the temperature decreases. The factors were that the mechanical properties of the material and the rheological properties of the lubricating medium changed strongly and nonlinearly at low ambient temperatures. In general, between $25^{\circ}C$ and $-10^{\circ}C$, the friction pairs of the two materials and ductile iron were comparable. The disc samples of spring steel and ductile iron wear more serious when the temperature dropped down $-10^{\circ}C$. Its ploughing was deeper and cracks appeared. According to the experimental results, the wear mechanism of the friction pair composed of two materials were summarized, which not only optimized the design of the hydraulic motor, but also provides reference data for establishing the online intelligent fault diagnosis system for marine hydraulic system.

This work presents a hybrid CMOS-RRAM integration of spiking nonvolatile computing-in-memory (nvCIM) processing engine (PE) that includes a 64Kb RRAM macro and a novel in situ nonlinear activation (ISNA) module. We integrate the computing controller and nonlinear activation function on-chip to compute convolutional or fully-connected neural network. ISNA merges A/D conversion and activation computation by leveraging its nonlinear working region. This eliminates the need for additional circuits to realize nonlinearity and reduces area by 43.7x w.r.t. the ADC scheme. The activation precision of ISNA can be configured from 1 to 8 bits to balance throughput, accuracy and power efficiency. The measurement of 4-layer LeNet shows such optimization improves 23.1% of computing speed via compromising a 2.5% relative accuracy drop. The proposed nvCIM PE achieves 16.9 TOPS/W power efficiency and a maximum spike frequency of 99.24 MHz.

Cognitive agents are typically utilized in autonomous systems for automated decision making. These systems interact at real time with their environment and are generally heavily power constrained. Thus, there is a strong need for a real time agent running on a low power platform. The agent examined is the Cognitively Enhanced Complex Event Processing (CECEP) architecture. This is an autonomous decision support tool that reasons like humans and enables enhanced agent-based decision-making. It has applications in a large variety of domains including autonomous systems, operations research, intelligence analysis, and data mining. One of the most time consuming and key components of CECEP is the mining of knowledge from a repository described as a Cognitive Domain Ontology (CDO). Given the number of possible solutions in the problems tasked to CDOs, determining the optimal solutions can be very time consuming. In this work we show how problems that are often solved using CDOs can be carried out using spiking neurons. Furthermore, this work discusses using the Intel Loihi manycore spiking neural network processor to solve CDOs using a technique inspired by a confabulation network. This work demonstrates the feasibility of implementing CDOs on embedded, low power, neuromorphic spiking hardware.

Traditional vehicle target detection algorithm requires selection of suitable features for different image and video scenes, which make poor generalization. This paper presents the mechanical identification of large-scale hydraulic engineering vehicle using SSD framework based on video big data. A large-scale hydraulic engineering vehicle uses image context information analysis to understand the detection target and predicts convolution characteristics of multiple scales. In the post-target detection process, we proposed a semantic window mining method to improve the recall rate, and we used a data-driven fuzzy object tracking method to realize accurate visual tracking of objects in video big data. The experimental results showed that the method has high recognition rate for hydraulic engineering vehicles, and has better accuracy and better effect than traditional target detection algorithms.

In order to improve the classification ability of the Twin Support Vector Machine (TWSVM), a new class of twice continuously differentiable piecewise smooth functions is used to smooth the objective function of unconstrained TWSVM and a class of smooth piecewise twin support vector machine (SPTWSVMd) is proposed. It is shown that the approximation accuracy and smoothness rank of piecewise functions can be as high as required. In order to reduce the influence of noise, the membership function is defined according to the distance between the sample points of each class and its intra-class hyperplane and a class of fuzzy SPTWSVMd (FSPTWSVMd) is proposed. The FSPTWSVMd can efficiently handle large scale and high dimensional problems based on the reduced kernel technique. The effectiveness of the proposed method is demonstrated via experiments on NDC datasets.

In this paper, we report our work on applying an artificial neural network (ANN) to interpret brain wave signals into words. Signals were acquired by a four-channel human electroencephalography (EEG) head set. EEG data were recorded through a video-guided user interface with time stamps. The objective of the experiment is to set up a two-word-based EEG library and to predict a random trail of participants' mental activities, in terms of words. We show that our algorithms can achieve a 90% prediction accuracy of the Bernoulli experiments.

© 2018 IEEE. Coupled matrix-tensor factorization has been suc- cessfully applied in various fields in the processing of coupled data. However, the unshared components between coupled data tend to make the joint decomposition inaccurate. In order to solve this problem, in this work, we propose a method to improve the traditional method by combining individual decomposition and coupled decomposition to analyze the shared and unshared components. Numerical experiments are given to illustrate the advantages of the proposed method compared to the existing approaches.

Some recent works revealed that deep neural networks (DNNs) are vulnerable to so-called adversarial attacks where input examples are intentionally perturbed to fool DNNs. In this work, we revisit the DNN training process that includes adversarial examples into the training dataset so as to improve DNN's resilience to adversarial attacks, namely, adversarial training. Our experiments show that different adversarial strengths, i.e., perturbation levels of adversarial examples, have different working zones to resist the attack. Based on the observation, we propose a multi-strength adversarial training method (MAT) that combines the adversarial training examples with different adversarial strengths to defend adversarial attacks. Two training structures - mixed MAT and parallel MAT - are developed to facilitate the tradeoffs between training time and memory occupation. Our results show that MAT can substantially minimize the accuracy degradation of deep learning systems to adversarial attacks on MNIST, CIFAR-10, CIFAR-100, and SVHN., 6 pages, 4 figures, 2 tables

With the advent of the Big Data era, large amount of high-dimensional data is generated, especially in the field of bioinformatics. In these data sets, they often contain many irrelevant, redundant and noisy features. Feature selection can reduce dimensionality and curtail the computational complexity during the analysis of classification problems. In this paper, we propose an improved binary quantum particle swarm optimization (iBQPSO) algorithm for feature selection. Firstly, the maximal information coefficient (MIC) is used to compute the correlation between feature and class label. Then, weak correlation features are removed. The optimal feature subset is selected by iBQPSO algorithm. Finally, the perfermance of our algorithm is measured by the accuracy of classification for the optimal feature subset. The experiments prove that the iBQPSO algorithm for feature selection can get better classification accuracy.

Crowbar circuit is the main means to achieve low voltage ride through (LVRT) in DFIG (doubly-fed induction generator). The fixed crowbar resistor is usually adopted in traditional crowbar circuit, which is difficult to meet the requirements of grid interconnection criterion for dynamic reactive power in the wind farm. This paper starts with the DFIG mathematical model with crowbar circuit. According to the analysis of the working state of the generator when the power grid is faulty, the analytical formula of the rotor current for the grid fault is obtained. Based on this, a self-adaptive crowbar control strategy is proposed. The adaptive crowbar resistance was set two times on the basis of limiting the rotor current. Then the theory of adaptive control strategy was verified by theoretical analysis and simulation experiments. The results show that the adaptive resistor control strategy combined with hardware and software increased the reactive power by about 0.9Mvar within 0.ls after grid fault occurs, which has a significant effect on optimizing the reactive power output of the DFIG's stator winding. And this adaptive resistor control strategy effectively improves its LVRT capability.

In this paper, we propose a novel framework to objectively evaluate the quality of movie trailers by fusing two sensing modalities: (1) Human Electroencephalogram (EEG), and (2) computer-vision based facial expression recognition. The EEG sensing data are acquired via a cap instrumented with a set of 4-channel EEG sensors from the OpenBCI Ganglion board. The facial expressions are captured while a user is watching a movie trailer using a regular webcam to help establish the context for EEG analysis. On their own, facial expressions reveal how engaged a user is while watching a movie trailer. Additionally, facial expression data help us identify situations where noises caused by muscle movement in EEG data. Using a shallow neural network, we classify facial expressions into two categories: positive and negative emotions. A quarter-central decision making strategy model is used to analyze EEG signals with a low pass filter activated by time stamp when large human movements are detected. A small human subject test showed that the adaptive analysis method can achieve higher accuracy than that obtained via EEG alone. Besides for movie trailer evaluation, this framework can be utilized in the future towards remote training evaluation, wearable device personalization, and assisting paralyzed people to communicate with others.

In this paper, we report our work on using machine learning techniques to predict back bending activity based on field data acquired in a local nursing home. The data are recorded by a privacy-aware compliance tracking system (PACTS). The objective of PACTS is to detect back-bending activities and issue real-time alerts to the participant when she bends her back excessively, which we hope could help the participant form good habits of using proper body mechanics when performing lifting/pulling tasks. We show that our algorithms can differentiate nursing staffs baseline and high-level bending activities by using human skeleton data without any expert rules.

Memristor with tunable non-volatile resistance offers in-memory computing capability that avoids the von-Neumann bottleneck. However, large-scale experimental demonstration to this end is yet to be implemented due to the immaturity of the device and integration technologies. Here in this paper we report our recent process in analog computing using analog-voltage-amplitude-vector input and analog-memristor-conductance matrix, with applications in signal and image processing. The vector matrix multiplication is processed in the memristor crossbars in one step, with 5–8 bit precision depending on the array size. The demonstration is made possible by high memristor yield (99.8%), stable multilevel memresistance states, linear current-voltage (IV) relation in the operation range, and low wire resistance between the cells.

Diffusive memristors with Ag active metal species are volatile threshold switches featuring spontaneous rupture of conduction channels at small electrical bias. The unique temporal dynamics of the conductance evolution originates from the underlying electrochemical and diffusive dynamics of the active metals in dielectrics, which can be explored for a variety of novel applications in unconventional computing. The superior I-V nonlinearity enables large crossbar arrays for high density non-volatile memories. The relaxation dynamics and the delay dynamics of the conductance evolution lead to faithful synaptic emulators and single-device threshold logic neurons, respectively. Unsupervised learning has been demonstrated with a fully memristive neural network consisting of these artificial synapses and neurons for the first time. In addition, the intrinsic stochasticity of the delay mechanism has been used to realize a true random number generators for security solutions.