Search

For face anti-spoofing, many methods have been proposed to improve the security of face recognition systems. Due to distribution discrepancies among different domains, it is difficult to seek a generalized space which can generalize well to unseen attacks. In this paper, we propose a framework based on meta-learning method to improve the generalization ability of face anti-spoofing. The feature extractor is trained with forcing the distribution of real faces more compact while the distribution of fake faces is more dispersed among domains. Then we add a hybrid-domain meta learner module to simulate multiple domain shift scenarios. Moreover, we add a refined triplet mining to constrain the distance between real faces and fake ones. Multiple gradient information is integrated to optimize the feature extractor and train the model with good generalization performance to unseen attacks of various scenarios. Extensive experiments on four public datasets show that our proposed method can get better generalization ability to unseen target domain compared with state-of-the-art methods. &nbsp

Accurate objects detection in remote sensing images is very important, because security, transportation, and rescue applications in military and civilian fields require fully analyzing and using these images. To address the problem that many small-sized objects in remote sensing images are difficult to detect, this paper proposes an improved S 2 ANET-SR model based on S 2 A-NET network. In this paper, the original and reduced image are fed to the detection network at the same time, and then a super-resolution enhancement module for the reduced image is designed to enhance the feature extraction of small objects, after that, the perceptual loss and texture matching loss is proposed as supervision. Extensional experiments are conducted to evaluate the performance on the general remote sensing dataset DOTA, and the results show that our proposed method can achieve 74.47% mAP, which is 0.79% better than the accuracy of S 2 A-NET.

Multiple sclerosis (MS) is a common chronic, autoimmune-mediated inflammatory and neurodegenerative disease of the central nervous system. The treatment of MS has enormous progress with disease-modifying drugs, but the complexity of the disease course and the clinical symptoms of MS requires personalized treatment and disease management, including non-pharmacological treatment. Transcranial magnetic stimulation (TMS) is a painless and non-invasive brain stimulation technique, which has been widely used in neurological diseases. In this review, we mainly focus on the progress of physiological assessment and treatment of TMS in MS.

To apply polarisation information of radar signal in radar reconnaissance equipment, a joint estimation algorithm of the radar signal direction of arrival (DOA) and polarisation parameters based on dual-polarisation antenna array has been researched in this study. The mathematical model of dual-polarised circular array has been given, and the joint estimation method of DOA and polarisation parameters for radar signal based on multiple signal classification (MUSIC) algorithm has been analysed. To overcome the shortcomings such as high calculation cost for the MUSIC algorithm, the DOA-matrix algorithm has been presented. DOA-matrix algorithm can effectively distinguish the radar signals with different polarisation states in the same direction by use of eigen-decomposition of the DOA-matrix. Through matrix correlation calculation, spectral peak search has been avoided, and the computational cost and algorithm complexity are greatly reduced. Finally, the feasibility of the DOA-matrix algorithm has been simulated and analysed. The simulation results show that the DOA-matrix algorithm can effectively complete the joint estimation of DOA and polarisation parameters for radar signal. The running time reaches millisecond level, which has a strong application value in engineering.

To solve the vibration suppression problem for a piezoelectric flexible cantilever beam with varying load, this study proposes a new type of multiple model hybrid adaptive vibration control method with a new multiple model switching cost index function. It employs a pre-filter consisting of secondary path and positive feedback path model to eliminate the positive feedback between the secondary actuator and reference sensor. By constructing a piezoelectric flexible cantilever beam as the controlled plant, we constructed a real-time control verification experiment platform based on a real-time MATLAB xPC target. We did a real-time control experiment comparison for vibration suppression of single-frequency, narrowband, and broadband disturbances. Experimental results show that the proposed multiple model hybrid adaptive vibration control algorithm is feasible and has good vibration suppression performance with rapid convergence speed.

An amendment to this paper has been published and can be accessed via the original article.

An ameliorated particle swarm optimization (APSO) algorithm was proposed in this paper to solve Bouc–Wen model parameter identification problem of piezoelectric actuator. Nonlinear dynamic acceleration coefficients and a modified particle position update approach are employed in PSO to improve the solution accuracy and accelerate the global convergence. Finally, Bouc-Wen model identified by APSO algorithm is experimentally verified on a micro-vibration control experimental platform and can converge to steady-state fast with FXVSSLMS algorithm.

Background: Sweetpotato root rot is a devastating disease caused by Fusarium solani that causes significant yield losses of sweetpotato in China. There is currently no effective method to control the disease. The breeding of resistant varieties is the most effective and economic way to control the disease. To date, quantitative trait locus (QTL) for resistance to root rot have not been reported and the biological mechanisms of resistance remain unclear in sweetpotato. Thus, it is necessary and worthwhile to identify resistance loci to help develop disease-resistant varieties. Results: In this study, we constructed genetic linkage maps of sweetpotato using a mapping population consisting of 300 individuals derived from a cross between Jizishu 1 and Longshu 9 by simple sequence repeat (SSR) markers, and mapped seven QTLs for resistance to root rot. In total, 484 and 573 polymorphic SSR markers were grouped into 90 linkage groups for Jizishu 1 and Longshu 9, respectively. The total map distance for Jizishu 1 was 3,974.24 cM, with an average marker distance of 8.23 cM. The total map distance for Longshu 9 was 5,163.35 cM, with an average marker distance of 9.01 cM. Five QTLs ( qRRM_1 , qRRM_2 , qRRM_3, qRRM_4 , and qRRM_5 ) were located in five linkage groups of Jizishu 1 map explaining 52.6-57.0% of the variation. Two QTLs ( qRRF_1 and qRRF_2 ) were mapped on two linkage groups of Longshu 9 explaining 57.6% and 53.6% of the variation. 71.4 % of the QTLs had a positive effect on the variation. Three of the seven QTLs, qRRM_3 , qRRF_1 , and qRRF_2 , were colocalized with markers IES43-5mt, IES68-6fs**, and IES108-1fs, respectively. Conclusions: To our knowledge, this is the first report on the construction of a genetic linkage map for purple sweetpotato (Jizishu 1) and the identification of QTLs associated with resistance to root rot in sweetpotato using SSR markers. These QTLs will have practical significance for the fine mapping of root rot resistance genes and play an important role in sweetpotato marker-assisted breeding.

Expressway traffic flow is an important traffic information of expressway. At present, the main way to detect expressway traffic flow is through video image collection, which is easily affected by external light blocking and has high calculation cost of image analysis. In order to solve the problems of poor accuracy of image analysis and large amount of computation in dynamic image analysis, based on the study of audio signal of expressway traffic, this paper proposed a method of expressway traffic flow status identification based on depth neural network. The method proposed in this paper firstly classifies and defines the traffic flow status, denoises the audio signal and extracts the features, then uses Deep Neural Network (DNN) to build the model to get the DNN model to identify the traffic flow status of expressway, and then pretrains the DNN model. Then the parameters of the model are optimized, and the Hidden Markov Model (HMM) is used to decode the model. Finally, the DNN model is used to estimate the observation probability of audio signals of different expressway traffic flow status, and the result of expressway traffic flow status identification is given according to the calculated probability. The accuracy of the proposed method is compared with that of the existing methods. Experimental results show that the proposed method performs better in recognition accuracy.

Stretchable sensors, as the important components of flexible electronic devices, have achieved progress in a variety of applications for monitoring physical or environmental conditions, such as sound, temperature, vibration, and pressure. However, it still remains a challenge to fabricate high performance stretchable humidity sensors. Herein, we present a novel stretchable humidity sensor, which was fabricated based on an ultrastretchable polyaniline composite fiber. Because of the composite fiber with a "twining spring" configuration (cotton fibers twining spirally around a polyurethane fiber) it maintains a stable electrical conductivity up to a strain of 200%. In addition, the conductivity of the composite fiber remains perfectly stable after 5000 cyclic stretching events of 200% strain. Incorporating the humidity sensitive properties of nanostructured polyaniline, the stretchable humidity sensor based on the composite fiber effectively maintains its humidity sensitivity at different elongations.

This study concerns adaptive feedforward and hybrid (combined feedback and feedforward) control systems for active vibration suppression of smart flexible beam with a tip mass. By analysing traditional feedforward filtered-X least mean square (FXLMS) algorithm and filtered-X variable step size least mean square (FX-VSSLMS) algorithm, a new hybrid FX-VSSLMS algorithm is developed to improve the control performance of feedforward controllers. A stainless steel beam with a tip mass is employed to simulate flexible robot arm with varying loads. Piezoelectric actuators and sensors are attached to the surface of the beam. Comparison experiments show the effectiveness and benefits of the proposed hybrid FX-VSSLMS method.

With the development of autonomous driving technology, it is of great practical and educational significance to consider the ethics in the process of its technology realization. On the one hand, compared with foreign countries, engineering ethics education still needs to be actively promoted and innovated in China. On the other hand, carrying out case study of engineering ethics meets the practical needs of training engineering talents. Based on this, this paper adopts a case study strategy centered on the autonomous driving ethics, and transforms the traditional teacher-centered teaching mode into a student-centered teaching mode in the form of role-playing and collaborative discussion. In addition, a curriculum design framework is proposed, which highlights initiative, participation and self-study guidance. Then, a simulation case of autonomous driving with increasingly complex scenarios is designed, guiding students to consider ethical issues in a deeper and more comprehensive way. As a result, the described teaching plan deepens the ethical awareness of engineering students in the project, and improves their ability to analyze and solve multi-dimensional engineering ethical problems.

With the continuous development and maturity of artificial intelligence, its application in many industries has gradually deepened, which has caused many ethical problems. This paper analyses the engineering ethical dilemma in the development of artificial intelligence from the aspects of intelligent robots, self-driving technology, killer robots and public information security, and further explores the causes of related problems. Finally, the corresponding countermeasures are put forward, including strengthening the supervision of artificial intelligence technology and products, enhancing the moral responsibility of scientists and engineers, strengthening international cooperation and building a human-machine fate community. The analysis of ethical problems and Countermeasures in artificial intelligence technology will help to solve the dilemma faced by AI technology in the development process to a certain extent.

Under the comprehensive background of the development of engineering education in the world today and the promotion of quality education in China, strengthening engineering ethics education of engineering masters meets the needs of engineering technology development, and is also an indispensable part of shaping high-quality engineers. However, engineering ethics awareness cannot be generated spontaneously. It needs to be cultivated and improved through engineering ethics education. Therefore, this paper designs a closed questionnaire from three aspects: school curriculum education, engineering ethics cognition and the application of ethical dilemma of driverless cars. The statistical data of the questionnaire shows that engineering masters do not pay enough attention to the engineering ethics curriculum and it is difficult to use the knowledge of engineering ethics to analyze and solve the actual ethical dilemmas. In addition, it is further found that there is a gap between professional masters and academic masters in the ethical awareness, and the reasons are related to the distribution of engineering ethics education courses offered by the school. This paper proposes three specific recommendations in response to the above problems.

The airborne laser weapon system is a kind of directional energy weapon system, which has a higher tracking accuracy requirement than other weapon systems, so that the traditional PID control cannot meet the requirements of high precision control. A compound axis servomechanism with preview model was used to build a complete simulation system, including detector model, fast mirror model and time delay model, etc. In order to improve the control accuracy, the active disturbance rejection control algorithm and the compound axis control structure were combined. The simulation system built in the article was verified by the function verification test of some transport disturbance data collected in practice. The results show that the tracking accuracy of the system is 5.16 μrad, which is 25 times of the traditional PID control. At the same time, a virtual battlefield scene was given, which was verified by the simulation system built in this paper. The tracking accuracy of pitch axis and yaw axis were both less than 10 μrad.

Endophytic actinobacteria that lived in any associations with plant tissues represented a rather unexplored area of actinobacteria compared with soils. Gynura cusimbua was a kind of medicinal plant which had prevention effects for high blood pressure, coronary heart disease, Alzheimer’s disease, atherosclerosis, etc. Endophytic actinobacteria of G. cusimbua might produce some secondary metabolites which had the same function as their host. Stem samples of G. cusimbua collected from Hainan Province were used to study their endophytic actinobacteria to find some new compounds. In order to avoid vast proportions of the host plant DNA in the metagenomic library, the strategies of enrichment of the microorganism cells after tissue digestion and exclusion of 16S rRNA gene derived from the plastid by digested with PvuII were used. Two sets of actinobacteria specific primers were used for targeting endophytic actinobacteria from metagenomic library. 63 positive clones of actinobacteria were selected for sequencing and constructing the phylogenetic tree of 16S rRNA gene, and the 16S rRNA gene sequence of 59 strains among them had higher similar to the closest type strain and belonged to Microbacterium, Arthrobacter, Micrococcus, Curtobacterium, Okibacterium, Quadrisphaera and Kineococcus, respectively. Others were in low similarity and belonged to unclassified Micrococcineae, unclassified Intrasporangiaceae and unclassified Microbacteriaceae.

The Energy Internet is a new type of energy utilization system based on power electronic technology and information and communication technologies. It uses energy and information to be closely coupled to achieve safe and efficient coordination and sharing. This paper analyzes the energy Internet basic concept and key technology, combined with the current energy demand of the Internet for information communication system and communication system problems, puts forward the hierarchical network design of energy Internet information communication.

Taking acidified Cd contaminated brown soil in Yantai as the research object, based on different exogenous salinity (NaCl) gradient (0%, 0.3%, 0.6%, 0.9%, 1.5%, 2% and 5%), indoor simulation experiments of Cd release were carried out after field investigation. Results showed that there was a significantly positive relation (r>0.90) between Cd release concentration/amount/ratio and exogenous salt (NaCl). Besides, the more exogenous salt (NaCl) was added; maximum release concentration/amount of Cd appeared the earlier. It was found that exogenous salt (NaCl) addition could obviously promote Cd release from acidified Cd contaminated brown soil. It was believed that this could be mainly due to the cation exchange between Cd2+ and Na+, together with the dissociation and/or complexation between Cl- and Cd2+. In addition, available adsorption sites reduction by exchange base in soil causing Cd changed from solid state to soil solution was also a probable reason.

An optimization method based on mode approach and improved genetic algorithms is developed for optimal FBG sensors placement in health monitoring and reconstruction of flexible thin-plate structure. The optimization objective is to obtain a kind of placement of FBG sensors which has the best reconstruction effects. The structure is modeled using Finite Element methods and modal analysis is carried out to create a modal space which provides data for both shape-reconstruction and error-calculation. The optimization is performed by improved genetic algorithm that uses the mean square error of reconstruction for the comparison of different optimal solution candidates. FBG sensors were arranged on the structural surface and experiment with an optimized sensor placement were performed to verify the effectiveness of the method and the simulation results.

It is of great significance to perform FBG sensor calibration in shape reconstruction for a board. It impacts on the shape reconstruction accuracy directly. Aiming at the deficiencies of traditional off-line calibration method, an online calibration for FBG networks based on improved particle swarm optimization algorithm was put forward, optimization model was established, and fitness function was given. The optimal goal is to obtain the minimum of fitness function. Improved particle swarm optimization algorithm was used to obtain the optimal solution, ie. strain-curvature coefficients of FBG networks. The simulation experiments verified the effectiveness of the method. Finally, experimental platform was built for experimental verification and analysis of the results of calibration. Experimental results showed that high reconstruction accuracy can be obtained by this calibration method, while the measurement characteristics of the FBG sensors can be reflected more accurately.

The control of a flexible transmission system under various conditions with various loads is an important challenge. This paper proposes a multiple-model adaptive control algorithm as a solution for the flexible transmission benchmark problem. Switching strategies and controller design methods are described in detail by means of the results of a stability analysis of the controller. To verify the effectiveness of the proposed algorithm, a flexible transmission experimental platform was employed to carry out comparison experiments using a robust controller, an adaptive controller, and the proposed controller. The experimental results showed that the proposed algorithm performed much better than the other controllers because in actual experiments, the other controllers failed under many conditions. The experimental parameter analysis of the proposed algorithm shows that the influence of parameters in the case of the proposed system is not significant when there is a sufficient number of models and reasonable dwell time. This means that the proposed algorithm can always guarantee stability control for flexible industrial transmission systems, and can quickly and stably track and adjust the target.

A transfer transistor structure achieving complete charge transfer in a CMOS image sensor pixel is proposed to eliminate image lag. This structure adopted double N implants for photodiode to ensure a wide channel of charge transfer, a surface P type preventing from dark current and especially a non-uniform doped poly-silicon gate to adjust the potential profile in channel. The simulation results show that if a light pulse of 10000lux illuminates for one integration cycle, the signal voltage of the following frame drops to 1/50000 of the first one. The electron density in N buried layer after charge transfer is around 108/cm3, far below the intrinsic value.

Control algorithm is one of the key elements for active vibration suppression of the space flexible structure based on smart materials. As least mean square (LMS) algorithm and recursive least square (RLS) algorithm are two fundamental algorithms for adaptive feed-forward filter algorithm drawing wide attention, this paper focuses on the algorithm process analysis and performance comparison of the two adaptive algorithms based on finite impulse response (FIR) structure. The design and related characteristic analysis of the controller is given as well as the realization process based on the theoretical analysis. Simulation is done using Matlab 7.0 to verify the correctness of the theoretical analysis. The simulation comparison between the two type algorithms shows that RLS algorithm has faster convergence and better control performance than LMS algorithm. On the basis of the simulation analysis, the actual vibration suppression experiment is done on the constructed experimental platform for piezoelectric flexible beam, and experiment result confirms the simulation effects.

As the accuracy of the control channel model directly affects the stability and the convergence of the active vibration controller for smart flexible structures, offline identification of the control channel model is inappropriate for a vibration system whose characteristic and parameter is time-varying. To solve the problem of the control channel model identification, an active vibration control algorithm based on online control channel identification is proposed in this paper. Adopting a noise signal as the input identification signal, the FIR filter structure as the system model structure, and FxLMS algorithm as the control algorithm, the detailed derivation of the algorithm is presented. To test and verify the feasibility and priority of the proposed algorithm, simulation analysis is done by Matlab, while the actual experiment is done on the experimental active vibration control platform. The experiment results show that the proposed algorithm is feasible with good control performance and rapid convergence.

Taking the piezoelectric flexible plate as research object, this paper presents an efficient method to simulate and analyze intelligent structures with MATLAB and ANSYS applied. Firstly, this paper focuses on the development of modeling method for intelligent structures with distributed sensors and actuators. Modal analysis and transient analysis were done by using ANSYS 11.0 software, with the finite model of the intelligent plate constructed. Then based on an APDL (the design language of ANSYS) program, the vibration control experiments were done. By analyzing different control performance of different actuator assign position, the optimization criterion suggested by NASA for piezoelectric sensors and actuators arrangement was verified. While the optimal configuration was chosen, the control channel and primary path model of the vibration control system were obtained with MATLAB system identification toolbox employed. Finally according to the identified model an iterative learning control system was developed. The analysis and experimental results show that the analysis method for intelligent structures proposed in this paper is economical and effective and the designed controller has a good vibration suppression performance.

Taking the piezoelectric flexible beam as the research object, this paper proposed a FXLMS algorithm based MIMO adaptive vibration control method to attenuate the oscillations of flexible structures at low frequency. By introducing the piezoelectric constitutive equations and analyzing the multi-channel adaptive filtered feedforward control algorithm and its FXLMS algorithm process, the model parameter identification strategy for the controlled structure as well as the controller design structure diagram are given in this paper. Structure modal characteristic analysis was done to determine the optimal configuration of piezoelectric sensors and actuators. Active vibration control experimental platform was established to verify the effectiveness of the FXLMS control scheme. The vibration control experiment was done after the control channel model was obtained. The experimental results show that the MIMO adaptive vibration controller proposed in this paper has a good control performance, with low order model and rapid convergence.

Taking the piezoelectric beam as research object, filtered-U least mean square (FULMS) control algorithm for feed-forward adaptive control is analyzed and compared with filtered-x least mean square (FXLMS) feed-forward control algorithm, which can solve the problem that the vibration feedback may affect the measuring of the reference signal in FXLMS. By introducing FXLMS and FULMS control algorithm, this paper gives the controller design structure, while structure modal analysis to determine the optimal placement of piezoelectric sensors and actuators and model parameter identification for the controlled structure are done. Then the active vibration control experimental platform is established to verify the effectiveness of FULMS control scheme as well as FXLMS control scheme. Meanwhile the experimental results of FXLMS and FULMS are compared. The performance comparison shows that the FULMS adaptive vibration controller has a better control performance, with more rapid convergence than FXLMS controller.

Vibration control of flexible structures is observed to be a good way to maintain structural integrity as well as to optimize performance for space structures currently. Taking the piezoelectric flexible structure as research object, this paper focuses on the methodology and implementation of a vibration suppression system. An efficient method to simulate and analyze the piezoelectric plate is presented with ANSYS applied, while an iterative learning control (ILC) algorithm is proposed as the control strategy. A finite element model of the flexible plate is constructed by an APDL (ANSYS Parameter Design Language) program. By transient analysis of piezoelectric flexible plate, the models of the control channel and the primary path are obtained. Finally, an ILC based controller, having good cancellation of the vibration of the flexible plate, is developed. The ANSYS experiments suggest that analyzing the piezoelectric flexible plate by ANSYS is a cost-effective method. The SIMULINK results show that the open-loop PID-type learning algorithm is effective for the active vibration control.

A dynamic range extension scheme applied to a time delay integration (TDI) CMOS image sensor (CIS) is presented. Two types of pixels with higher and lower conversion gain are adopted in the pixel array, which are suitable for capturing images in low and high illumination respectively. By fusing the two kinds of pixels' output signals in the process of TDI accumulation, a high dynamic range image can be achieved. Compared with the traditional multiple integration technique, no photoelectrons generated during the exposure time are discarded by the reset operation, and thus a higher level of signal-to-noise ratio (SNR) can be retained. A prototype chip with an 8 × 8 pixel array is implemented in a 0.18 μm CIS process, and the pixel size is 15 × 15 μm2. Test results show that a 76 dB dynamic range can be achieved in 8-stage TDI mode, when the SNR boost can reach 7.26 dB at 90.8 lux.

To improve the full well capacity (FWC) of a small size backside illuminated (BSI) CMOS image sensor (CIS), the effect of photodiode capacitance (CPD) on FWC is studied, and a reformed pinned photodiode (PPD) structure is proposed. Two procedures are implemented for the optimization. The first is to form a varying doping concentration and depth stretched new N region, which is implemented by an additional higher-energy and lower-dose N type implant beneath the original N region. The FWC of this structure is increased by extending the side wall junctions in the substrate. Secondly, in order to help the enlarged well capacity achieve full depletion, two step P-type implants with different implant energies are introduced to form a P-type insertion region in the interior of the stretched N region. This vertical inserted P region guarantees that the proposed new PD structure achieves full depletion in the reset period. The simulation results show that the FWC can be improved from 1289e− to 6390e−, and this improvement does not sacrifice any image lag performance. Additionally, quantum efficiency (QE) is enhanced in the full wavelength range, especially 6.3% at 520 nm wavelength. This technique can not only be used in such BSI structures, but also adopted in an FSI pixel with any photodiode-type readout scheme.

The charge transfer efficiency improvement method is introduced by optimizing the electrical potential distribution under the transfer gate along the charge transfer path. A non-uniform doped transfer transistor channel is introduced to provide an ascending electrical potential gradient in the transfer transistor channel. With the adjustments to the overlap length between the R1 region and the transfer gate, the doping dose of the R1 region, and the overlap length between the anti-punch-through (APT) implantations and transfer gate, the potential barrier and potential pocket in the connecting region of transfer transistor channel and the pinned photodiode (PPD) are reduced to improve the electrical potential connection. The simulation results show that the percentage of residual charges to total charges drops from 1/104 to 1/107, and the transfer time is reduced from 500 to 110 ns. This means the charge transfer efficiency is improved.

This paper has made a study of the scheduling and the allocation of the byproduct gases in a large scale steel works. In order to realize the optimal utilization and allocation of these gases with a micro-computer, the paper introduces the model of multi-objective optimization and proposed a new decomposition-coordination method, called Constraint coordination method, which is simpler and more effective than Dantzig Wolfe decomposition method. In the last part of tho paper, the method has been successfully used in the computation of optimal allocation of the gas resources in the steel works.