Search

In this paper, we are interesting in inferring 3D pose estimation of aircraft object leveraging 2D key-points localization. Monocular vision based pose estimation for aircraft can be widely utilized in airspace tasks like flight control system, air traffic management, autonomous navigation and air defense system. Nonetheless, prior methods using directly regression or classification can not meet the requirements of high precision in aircraft pose estimation context, other approaches using PnP algorithms that need additional information such as template 3D model or depth as prior knowledge. These methods do not exploit to full advantage the correlation information between 2D key-points and 3D pose. In this paper, we present a multi-branch network, named AirPose network, using convolutional neural network to address 3D pose estimation based on 2D key-points information. In the meantime, a novel feature fusion method is explored to enable orientation estimation branch adequately exploit key-points information. Our feature fusion method significantly decreases 3D pose estimation error also avoids the involvement of RANSAC based PnP algorithms. To address the problem that there is no available dedicated aircraft 3D pose dataset for training and testing, we build a visual simulation platform on Unreal Engine 4 applying domain randomization (DR) skill, named AKO platform, which generates aircraft images automatically labeled with 3D orientation and key-points location. The dataset is called AKO dataset. We implement a series of ablation experiments to evaluate our framework for aircraft object detection, key-points localization and orientation estimation on AKO dataset. Experiments show that our proposed AirPose network leveraging AKO dataset can achieve convincing results for each of the tasks.

EHB (electro-hydraulic brake system) can provide active and decoupled brake for electric intelligent vehicles. However, modelling and control of automotive electro-hydraulic brake systems are always challenging due to the complex and highly nonlinear dynamic characteristics of the braking process. In this paper, a nonlinear backstepping control algorithm for EHB based on the bond graph model is introduced. A nonlinear, single-wheel brake system model is constructed using the bond graph method, in which the capacitive effect of the hydraulic fluid and the damping and inertia effects of calipers are considered. Based on this nonlinear model, a backstepping controller with good robustness is designed, and the control signals of the inlet and outlet valves are described by a unified expression. Moreover, the effects of the controller parameters on brake-pressure control are discussed in details, and suitable design parameters are chosen. Finally, the proposed algorithm is verified by the simulation and hardware-in-the-loop tests. The results show that valves are properly regulated by the backstepping controller, and the brake pressure follows the desired value rapidly and accurately.

This paper proposes a novel gyro drift estimation method for the inertial navigation system (INS), which introduces the particle swarm optimization (PSO) algorithm into the error estimation problem. PSO-based estimation model is established. The position error and velocity error of INS are considered as the performance criterions of the PSO fitness function. Compared with the traditional gyro drift estimation methods, the advantages or contributions of the proposed method can be summarized as follows: 1) the proposed method does not require any prior information about inertial sensor error or observation noise; 2) particular motion for the carrier of INS is not needed, and; 3) the external information provided by other navigation systems could be discontinuous. The simulation experiments and field tests are performed, which validate the efficacy of the proposed method.

This paper studies the design of adaptive event-triggered protocols for consensus of linear multi-agent systems (MASs) with external disturbances. Different from most of the existing results that deal with undirected graphs, this paper addresses directed graphs, specifically graphs that are assumed to be strongly connected. Inspired by a recent development [1], this paper devises novel adaptive event-triggered protocols for linear MASs with all agents subject to external disturbances. Two specific designs of composite triggering conditions are analysed and discussed. Compared with the disturbance-free case, additional constraints need to be introduced to deal with external disturbances and moreover the time-dependent terms are allowed to take any finite functions, rather than a class of $L_{1}$ functions.

At present, in the investigation stage, public security organs, procuratorial organs and people's courts often need to conduct electronic data forensics analysis on criminals' WeChat and other social software to obtain sensitive information. This paper proposes a forensic analysis method based on chat history data of Wechat on Android smart phone. By using an Improved Naive Bayes Algorithm with redefinable the weights of the feature words, an improved Naive Bayes Classifier (NBC) model is established. The classifier is used to classify the chatting texts in WeChat, in order to find the sensitive information in the chatting texts efficiently. Experiments show that this model can improve the efficiency of acquiring sensitive information in WeChat, and has obvious effects on saving evidence collection time and improving the efficiency of criminal detection.

Fractional repetition (FR) codes are a class of repair efficient erasure codes that can recover a failed storage node with both optimal repair bandwidth and complexity. In this paper, we study the minimum distance of FR codes, which is the smallest number of nodes whose failure leads to the unrecoverable loss of the stored file. We consider upper bounds on the minimum distance and present several families of explicit FR codes attaining these bounds. The optimal constructions are derived from regular graphs and combinatorial designs, respectively.

A new geometrical model of divisional rough surface is established for simulating the surface of mountainous region. The Gaussian spectrum is applied to represent the statistic properties of mountainous surface. The modeling theory of divisional rough surface is introduced. Two modeling methods for constructing the borderline are given. what's more, the iterative physical optics (IPO) is utilized to obtain the monostatic radar cross-section (RCS) of the divisional rough surface. Several unique phenomena and useful conclusions can be obtained, which can be widely used in remote sensing simulations of mountainous region.

The melody extraction of polyphonic music is always an important research topic. Most melody extraction algorithms are based on computing a saliency function of pitch candidates or separating the melody source from the mixture. With the popularity of deep learning, data-driven methods based on deep neural networks, are gaining more and more attentions in the research of melody extraction. In this paper, our data-driven approach based on deep learning to treat melody extraction as a pixel-wise multi-task classification problem in image processing. The proposed model is composed of the advanced semantic segmentation Attention U-net that predicts the pitch salience map of the singing melody and voice detection model that estimate a voice activity of the polyphonic music and use a Softmax and an Argmax layer to get the final melody estimation. Experiments show that the proposed method can achieve better experimental results with small training dataset.

Off-shore wind turbine is a new hot spot in the field of new energy. Due to the characteristics of high salt fog, high humidity and long sunshine time in the marine environment, the anti-corrosion performance of air radiator in the water cooling system of high-power wind turbine has higher requirements. By analyzing the air radiator core used in high-power off-shore wind turbine project, this paper analyzes the effect of the existing anti-corrosion technology and corrosion mechanism, and puts forward the anti-corrosion design method and treatment method of the air radiator, which can provide guidance experience for related engineering projects.

With the promotion of technology and policy, wind power generation is developing towards large-scale and offshore. Due to the great amount of heat and high maintenance cost, higher requirements are put forward for the cooling system, among which the vibration is particularly important. In this paper, the vibration problem of the wind turbine in the generator cooling system is studied. The vibration experiment is carried out and found that the vibration can be effectively controlled only when the motor frequency is dynamically balanced in all the range; Strengthening the connection between the wind turbine and the bracket, the vibration velocity is reduced from 48.1mm/s to 12.4mm/s; The frequency corresponding to the peak vibration velocity is changed at the same time; According to the modal analysis, the range of the natural frequency of the wind turbine coincides with that of the motor frequency, causing the resonance. In the structural design of the wind turbine, the coincidence of natural frequency and motor frequency should be avoided; The simulation results are in agreement with experimental results. This two methods can effectively solve the vibration problem.

This paper addresses the problem of a stratospheric airship with external disturbance circumnavigates a moving target on a plane with time-varying velocity. An algorithm of moving target circumnavigation is proposed based on the line-of-sight guidance law. First, the target enclosing controller is proposed, which can navigate the stratospheric airship surrounding a moving target at a given distance by calculating the desired yaw rate and velocity. Second, an adaptive backstepping attitude controller to track the desired yaw rate is introduced by estimating the boundaries of dynamic couplings and unknown disturbance. Finally, a velocity controller is added to track the desired speed. Numerical simulations show the validity of this algorithm.

Consumer purchase prediction is of great significance for reducing marketing costs and improving return on investment of companies. Recently, spatial-temporal data mining has aroused increasing concern. In this paper, we propose a hybrid deep learning model (EE-CNN) for purchase prediction, which combines entity embedding and convolutional neural networks. In empirical experiments, we first explore the purchase location pattern of different consumer groups on data sets from a retail company of China. After that, our proposed EE-CNN model is utilized to predict consumer purchase behavior. It turns out that location data can help improve the performance of purchase prediction models in general. Meanwhile, our proposed EE-CNN model outperforms baselines used in the experiments. Our research provides significant guidelines for the marketing decisions of enterprise marketers.

In the era of big data, the availability of large amounts of spatio-temporal data provides a new path for customer analytics. However, the related research in the retail industry is underdeveloped. In this paper, we introduce spatio-temporal data mining into retail customer analytics and conduct experiments on large real-world data sets from a retail company containing millions of customer purchase records. Spatio-temporal clustering and a new hybrid sequential pattern mining method are used to discover the characteristics of customer behavior at the aggregation as well as the individual level. The typical spatial and temporal distribution of customers and main customer clusters are obtained. Some interesting sequential purchase patterns are also found. Our research will provide not only a new analytic framework for academia but also some guidelines for better development of marketing strategies in the retail industry.

Accurate preoperative differential diagnosis of fat-poor angiomyolipoma (fp-AML) and clear cell renal cell carcinoma (ccRCC) is essential for proper treatment planning. In this paper, we develop an effective radiomics model for reliable noninvasive discrimination of fp-AML from ccRCC, which incorporates a sophisticated feature selection procedure and a sparse radial basis function neural network (sRBFNN). Specifically, 774 three-dimensional radiomics features are first extracted from contrast-enhanced computed tomography (CECT) images. Pearson’s correlation matrices and Welch’s t-test are used to remove the insignificant features, then sequential forward floating selection method is utilized to select the discriminative features. Finally, the sRBFNN is employed for classification. The radiomics model is examined by leave-one-out cross validation and yields the best performance of 90.00%, 66.67%, 100.0%, and 0.9173 in terms of prediction accuracy, sensitivity, specificity, and the area under the receiver operating characteristic curves, respectively. Furthermore, the reliability of the predictions is verified by evaluating the probabilistic outputs of the sRBFNN. The experimental results demonstrate that the developed radiomics model has great potential in the noninvasive discrimination of fp-AML from ccRCC.

Regenerating codes (RGCs) have recently been proposed to reduce the repair traffic of ( $n,k$ ) erasure-coded distributed storage systems. Moreover, RGCs can also be used in a scalable distributed storage scenario where $n$ is increased (decreased) to upgrade (degrade) redundancy while maintaining the maximum distance separable property of erasure codes. In this paper, we propose a new application of minimum storage regenerating (MSR) codes in storage scalability. The connection between repairing invalid nodes and adding new nodes suggests that the two processes can be unified in the same framework. We consider both single and multiple node situations, and two methods for constructing multiple nodes are proposed: concurrent and sequential. We focus on proving the achieved capability of concurrent MSR that can consume minimum traffic for generating multiple nodes. Because concurrent MSR is sensitive to both the number of helpers and added nodes, sequential methods make scalable MSR generalizable. The examples show that the scalable MSR codes have the same advantage of saving network traffic as repairing failures.

Fractional repetition (FR) codes form a special class of minimum bandwidth regenerating codes by providing uncoded repairs with a table-based repair model. In this paper, we focus on a lower bound on the reconstruction degree of FR codes, which is the smallest number of storage nodes required for data retrieval. We show that for an FR code with reconstruction degree attaining this lower bound, the corresponding dual FR code is optimal with respect to an upper bound on the file size, and vice versa. Using this duality relationship, we present several families of FR codes with optimal reconstruction degree.

This paper addresses a finite-time autonomous shipboard landing problem of a model helicopter subject to output constraints, parametric uncertainties and external disturbances. By establishing the relative motion model between the helicopter and the ship, the shipboard landing problem is converted from a general trajectory tracking problem to a stabilization problem. Due to the under-actuated property of the helicopter, the whole process of the autonomous shipboard is divided into two steps. The helicopter is first commanded by a relative position controller (RPC) to reach above the ship. As it reaches, a relative attitude-altitude controller (RAC) is initiated to guide the helicopter to descend steadily on the ship. RPC and RAC are both devised in terms of the tan-type Barrier Lyapunov Function theory, with an adaptive estimation method inserted to compensate for the impact of uncertainties and disturbances. It is demonstrated from the stability analysis that the proposed control strategy can ensure that system tracking errors converge to small sets around zero in the finite time and never violating constraint requirements. Numerical simulations are implemented to further validate the remarkable control performance.

In most textbooks on principles of automatic control, very limited content is given on the angles of arrival or departure from breakaway points of root loci. Sometimes, for simplicity of description, the angles of arrival or departure from breakaway points are given by $180^{\circ}/ l$ apart, where l denotes the number of branches of root loci that meet at the breakaway point; however, no detailed proof is given for this result. In this paper, detailed proof is provided for the angles apart of the root loci branches from the breakaway point. It is proved that, the result is valid with strictly positive real open loop transfer function in both cases of unit negative feedback and unit positive feedback. The proof is based on the property of angles of departure or arrival and the angle condition of the root locus itself. Several examples are given to support the proposed proof. The proposed results can be lectured to students for better understanding and more precise plot of root locus.

A new unconstrained distributed model predictive control (DMPC) is extended for double-integrator systems with switching interaction graph. The task is to stabilize a static formation in the leader-follower framework. The proposed DMPC is implemented in a receding horizon way, where the predicted own control and neighboring controls are calculated as decision variables of optimization, but only the own optimized current control is exerted. Sufficient conditions are provided, such that the static formation can be stabilized asymptotically by the proposed DMPC in case of switching interaction graph.

In this paper, a novel planar path-following controller is presented based on the implicit path formulation. In the design, the path-following problem is restated as a stabilization problem of a second order system, i.e., designing the forward speed and the steering law to stabilize the heading error and the lateral distance to zero. Based on a new interesting design of the sliding surface, the proposed controller achieves the global asymptotical stability of the heading error and the implicit lateral distance at the origin. Both simulation and experimental results are presented to validate the effectiveness of the proposed controller.

This paper presents a torque ripple reduction method using optimized combination of positive voltage vectors (PVV) and new zero voltage vectors (ZVV) in direct torque control (DTC) of brushless DC motor (BLDCM). The switching table is optimized based on the analysis of duty ratio modulation to get the optimized combination of PVV and ZVV. Comparison between different control methods has been carried out by numerical simulation and the proposed method is validated quite effective to reduce torque ripple.

We present a joint-detection quantum receiver scheme. Performance of the receiver is improved by adaptive control strategy based on Bayes criterion and optimized displacement followed by single photon detection in each temporal slot. The adaptive quantum receiver is applied to on-off keying (OOK) and binary phase shift keying (BPSK) modulated binary coding signals. Numerical simulation shows that the adaptive quantum receiver can outperform the standard quantum limit (SQL).

The problems of a spacecraft attitude tracking and disturbance rejection are considered. Attitude tracking dynamics are formulated with typical Euler-Lagrange form, and a nonlinear controller is designed by combining a passivity-based adaptive method with an internal model. The unknown constant disturbance and inertia uncertainties are estimated online by element-wise adaptive update laws, while the unknown periodic disturbance is compensated by the internal model. Asymptotic stability of the closed-loop system is proved in the Lyapunov framework. A numerical example is also given to validate the theoretical results of this work.

This paper investigates the problem of attitude tracking control with prescribed tracking performance for a spacecraft exposed to unknown inertial matrix and actuator failure. Based on the prescribed tracking performance technique, an adaptive fault-tolerant controller is adopted by taking backstepping control and neural-network into consideration. By estimating the bounds of time-varying parameters induced by fault uncertainties, the influence of actuator faults is successfully compensated for. By estimating the norm of the optimal weight vector of neural networks, there is only one parameter needs to be estimated, which significantly reduces the computational burden. It is proved that the proposed controller can guarantee that all closed-loop signals are bounded and the tracking errors are converged into an arbitrarily small residual set with prescribed performance. Simulation results are presented to demonstrate the effectiveness of the proposed control scheme.

A model predictive controller (MPC) is proposed in this paper for a 3-DOF flapping-wing unmanned aerial vehicle (UAV) to track the reference set point. The proposed MPC is designed by using an averaged model of the 3-DOF flapping UAV, where the control constraints are considered. In the formulation of the MPC, nonlinear control inputs are transformed into linear ones, and constraints are re-designed for the linear control inputs, such that original constraints are satisfied. With the proposed MPC strategy, the set-point tracking error of the flapping-wing UAV can be proved ultimately bounded. A numerical example is provided to demonstrate the theoretical result.

This paper presents a robust optimal flight control approach for a class of tail-sitter unmanned aerial vehicles (UAV s) to achieve the optimal tracking control performances and robustness properties in the flight mode transitions. It is proven that the attitude tracking errors can converge into a given neighborhood of the origin in a finite time. Simulation results are presented to show the effectiveness of the proposed robust optimal control method.

Consumer credit scoring is an essential part of credit risk management in the fast-growing consumer finance industry and various data mining techniques have been proposed and used on it. Recently, deep learning techniques have gained significant popularity and shown excellent performance in many fields such as image recognition, computer vision and so on. In this paper, we try to take the advantage of deep learning and introduce it into consumer credit scoring. We propose a hybrid model that combines the well-known convolutional neural network with the feature selection algorithm Relief. Experiments are carried on a real-world dataset from a Chinese consumer finance company, and the results show that the proposed model gets superior performance in comparison with other benchmark models such as logistic regression and random forest.

To alleviate the problem of “topic drift” in hot topic on web, the hot topic detection system is implemented. Our system contains the parts as follow. First, news data are collected from Web and cut into the Chinese word segmentation; second, the word weight is calculated by the improved word frequency-inverse document frequency formula (TF-IDF); at last, the Single-Pass clustering algorithm is used for text clustering. The TF-IDF formula is improved, which contains the position information of the word. Experiments show that the hot topic can be effectively identified by this system.

A high-level dynamic mission allocation strategy is presented for a multi-UAV system to intercept multiple intruding targets. The objective of the dynamic mission allocation is to optimize the time-efficiency of the multiple interceptions. In the dynamic mission allocation, the optimization is carried out repeatedly in discrete-time before interceptions are completed, such that the outcome of mission allocation is improved based on feedback of real-time information. The performance of the dynamic mission allocation is proved to be superior (or at least equal) to that of a previously proposed static mission allocation.

More and more Advanced Driver Assistance Systems (ADAS) are entering the market for improving both driving safety and comfort. To improve the system performance, in particular, the acceptance and adaption of ADAS to human drivers, it is important to understand human drivers' driving habits that make the systems more human-like or personalized for ADAS. The research presented in this paper proposes a method to classify and recognize drivers' driving styles. A typical testing scenarios is created under a real-time Driver-In-the-Loop Intelligent Simulation Platform (DILISP) with both PanoSim-RT® and dSPACE®. The driving styles are defined and classified into three categories based on the root mean square of vehicle acceleration, the Time-to-Start and the Time-To-Follow. Samples with 21 drivers are used for the testing with driving data collected, analyzed and further employed for driving style recognition via Multi-dimension Gaussian Hidden Markov Process (MGHMP). Test results show that driving styles can be classified clearly and identified effectively using the proposed classification and identification strategy.