Abstract:As one of the current research hotspots in artificial intelligence, deep learning has received widespread attention. With the powerful feature representation and learning capability, the deep learning has increasingly become the technical basis for the development of intelligence in the military field. In combination with the latest development of deep learning, this paper points out that the rapid development of deep learning is attributed to the breakthrough of theory, the remarkable improvement of computer computing power and the widespread popularity of opensource software, focuses on combing the current main deep learning hardware platforms and programming frameworks, and summarizes the characteristics and research progress of each. Further, the application and shortcomings of the deep learning in typical military areas such as target recognition, situational awareness and command decision are summarized. Finally, the challenges faced by military application of deep learning are analyzed, including difficulties in data acquisition, insufficient ability to handle uncertain and incomplete information and multidomain information, low accuracy and realtime, and poor interpretability and comprehensibility. In view of such challenges mentionedabove, the paper looks forward to the future, the possible development directions, and the trend.

Abstract:Aimed at the problems that existing antenna size is excessively dependent on material properties, there are in contradictions between antenna stealth and radiation performance, and the antenna beam reconfigurable method and the performance are limited, this paper mainly discusses the antenna enabling technology based on field enhancement, the field confinement and slow wave effect of spoof surface plasmon polarition (SSPPs), including the antenna miniaturization technology, the radar cross section (RCS) suppression technology, and the beam reconfigurable technology. And the abovementioned provides the research of novel antennas with innovative ideas and options.

Abstract:In view of the practical needs of low outofband scattering of antenna, through adjusting the response thickness of polarized conversion metasurface by combining the frequency selective surface, a metasurface is designed to utilize both efficient crosspolarized reflection and copolarized wave transmission window. Furthermore, this paper utilizes the geometric phase in response only to crosspolarized waves, through the checkerboard array configuration, for realizing the low scattering characteristics on both sides of the translucent window, in which the penetration loss of 10.0~11.0 GHz insertion loss is less than 1 dB and up to 0.8 dB. Besides, and the Low scattering characteristics are achieved in the 4.5~7.5 GHz and 12.0~17.0 GHz bands with RCS reduction of nearly 10 dB. The simulation and experimental results show that the proposed metasurface design method and architecture are characterized by the broadband RCS reduction and low penetration loss, and have potential application value in stealth radome.

Abstract:This paper proposes a method of combining the metasurface with antenna array to realize highgain antenna array with simultaneous low backward scatterings. The metasurface composed of multilayered structure is characterized by allowing the xpolarized wave to transmit the ypolarized wave to reflect. Therefore, the radiation from an xpolarized antenna array can pass through the metasurface, and at the same time, diffusion scattering occurs when the outofphase metaatoms are used to form the checkerboard or random arrangement. The simulated results show that a peak gain of 18.8 dB and an operating band of around 12.5% can be obtained by the method. When the finding wave is xpolarization, the RCS of the antenna reduces by 10 dB from 9.5 GHz to 10.1 GHz and the peak reduction is 25 dB, covering the operating band of the antenna. For ypolarized incidence, 10 dB RCS reduction can be achieved from 7.8 GHz to 12 GHz, and the peak reduction is over 35 dB. The measured results are identical with the simulated results. The proposed design method can reduce the RCS effectively by using the transmissionreflectionselective metasurface, and the abovementioned provides lowscattering antenna design with a new path.

Abstract:Metamaterials demonstrate a series of unique properties in which natural materials do not have, and these properties have powerful manipulations to electromagnetic field, acoustic field and other physical fields. Being capable of absorbing sound wave and electromagnetic wave in broadband range, a multifunctional metamaterial composed of the multisized Helmholtz resonators and multisized metal resonators, is designed. The results show that the device could achieve an acoustic coefficient of more than 0.8 in the 690~927 Hz region and a microwave absorption over 80% in the 9.11~11.10 GHz range. The multifunctional metamaterials proposed are capable of absorbing both acoustic and electromagnetic waves with broadband and high efficiency and have potential applications in the aspects of noise pollution prevention and electromagnetic protection.

Abstract:In view of the limitation of traditional plasma optical sensors, and being capable of enhancement of absorbing infrared waves by magnetic resonance, a multiresonant nonplasma optical sensor based on black phosphorus (BP) is proposed. A BP monolayer and a simple dielectric structure are adopted by the sensor, Due to the anisotropy of BP, the sensor can achieve different sensitivities and figure of merits (FOMs) in different crystal directions. Through optimizing the sensor structure, two resonant peaks which are very narrow are obtained in different directions of BP crystals around 8 μm wavelength, and the sensitivity of the sensor is improved by the relatively low full width at half maximum. The highest absorption rate reaches up to 99.6%, while the highest sensitivity reaches up to 180 nm/RIU and FOM reaches up to 261 nm/RIU.The simulation results are verified by the coupled mode theory, and the physical mechanism is revealed. The sensor structure has great potentials for the future optical sensing application.

Abstract:With FSS structure being prepared in the ceramicbased wedge cavity, a conductive silver paste coating process is adopted to prepare a uniform metal coating on the inner surface of the quartz fiber reinforced silica composite material (SiO2f/SiO2) wedge, and then fine structure patterns on the metal coating are etched by performing the inner surface Laser processing technology and its performance is characterized. The results show that the thickness of the metal coating is 12.08 μm, the conductivity at normal temperature and 800 ℃ is 1.57×107 S/m and 1.496×107 S/m respectively, the adhesion reaches 0 level, and the dimensional accuracy of the microstructure is 0.193 2 mm.This process can realize the preparation of a frequency selective surface on the inner wall of the wedge with a large aspect ratio, the subsequence can be applied to the preparation of the FSS structure of the inner surface of radome, antenna window and other products.

Abstract:Gliding arc discharge plasma has obvious advantages in the aspects of broadening the ignition and extinction boundary of combustor, reducing ignition delay time and improving combustion efficiency. On the basis of exiting technology of fuel pyrolysis dome, this paper optimizes the position of gliding arc discharge and innovatively develops the pilot flame dome based on gliding arc discharge plasma actuation. And the experimental study on discharge characteristics is carried out to investigate the arc dynamic characteristics, gliding mode, average breakdown voltage, power under conditions of different airflow rates and input voltages. The results show that the Pilot flame dome can form a stable rotating gliding arc discharge area between Venturi tube and fuel nozzle, and there are two different discharge modes, i.e. Steady Arc Gliding (A-G) mode and Breakdown Gliding (B-G) mode, which are significantly affected by the airflow rate and input voltage. When the airflow rate is less than 200 L/min, the discharge is mainly in A-G mode at the input voltage of 140~240 V, and develop to B-G mode with the increase of airflow rate. However, the proportion of A-G mode increases gradually with the increase of input voltage. The average breakdown voltage and rotating angular velocity decrease with the increase of input voltage or the decrease of air flow, but the average discharge power increases with the increase of input voltage.

Abstract:Aimed at the problems that the aircraft structure load measurement is affected by the structure temperature in the process of flight test of aircraft structure load measurement in recent years, on the basis of analyzing the cause and mechanism of thermal output of straingage bridge for load measurement, a method is proposed based on polynomial fitting for analysis and correction of thermal output of straingage bridge. This method is valid through ground temperature being tested. The method is used to correct the measurement results of several aircraft structural load flight tests, and a good correction effect is obtained, effectively improving the accuracy of load measurement. The research results provide the correction of thermal output of the straingage bridge in load measurement with an applicable theoretical basis and engineering method, and provide an important technical support for flight test and verification of aircraft structure load.

Abstract:In the process of operation of a twinrotor turbofan engine, if inflection point of engine control plan changes, this may affect the engine performance. In order to study the influence of aeroengine performance degradation on the inflection point of engine control plan, a component level model is modified by component characteristics. A personalized degradation model of an aeroengine is built, and the effects of control plan inflection point advance and major component decline on engine performance are analyzed. The decline of engine components is studied, and the change of exhaust temperature is analyzed. The research result shows that with the deterioration of turbofan engine components, the exhaust temperature increases, and this makes the inflection point of engine control plan moving forward. And, the inflection point of engine control plan moves forward greatly after the engine performance declines. The forward movement of the inflection point of the control plan causes the thrust loss of the engine at the maximum state to reach 3.65% of the total state of the engine when the engine is not in recession. The thrust is reduced by 0.5% compared with engine degradation on no consideration of the change of control plan.

Abstract:In order to find out the causes of two abnormal liftoff acceleration fastturning accidents occurring in succession, according to the mechanism of the helicopter tail rotor vortex ring, combination with the state characteristics of the accident helicopter, the calculation and analysis of the tail rotor vortex ring characteristics are carried out, and the results of the parameter calculation are consistent with the data recorded by the accident helicopter. The research result shows that this type helicopter being in the small pull state of the tail rotor on the ground, and the critical deflection angle speed of the corresponding tail rotor vortex ring being small, once an unexpected deflection occurs to the abnormal liftoff, there is in a situation of easily entering into the tail rotor vortex ring state and hardly changing out , i.e. the characteristics of the tail rotor vortex ring going from bad to worse. After analysis, the relevant measures are put forward to improve the safety level of this type helicopter takeoff and landing. The findings also refer to other similar helicopter.

Abstract:According to the special rotor structure of permanent magnet toroidal motor (PMTM), an electromagnetic parameter equation of the motor is established. The time varying mathematical model with the rotor position information of planetary gear is derived. The periodic fluctuation characteristics of PMTM are analyzed by using the statespace method. In combination with the timevarying parameters of PMTM, the setting parameter of PI controller is made. In combination with the parameter perturbation caused by the rotation of planetary wheel, the rotor position information obtained from the closedloop system feedback is introduced into the terminal sliding mode controller, enabling terminal sliding mode control parameters to track the timevarying parameters of toroidal motor. The simulation results of terminal sliding mode control and PI control are compared and analyzed, and the results show that the terminal sliding mode control strategy weakens effectively the fluctuation amplitude of output torque of PMTM, and is good at response speed, in interference immunity, and in robustness.

Abstract:In view of the abnormal situation in the measurement, an Adaptive Robust Squareroot ContinuousDiscrete Cubature Kalman Filter algorithm is proposed based on the Mestimated. The algorithm is that the target tracking problem is modeled as a continuousdiscrete time model, the idea of improved M estimation is integrated into the continuousdiscrete cubature Kalman filter algorithm, threshold abnormal measurements are made by using Mahalanobis distance, a correction factor is introduced, and the size of the observed noise covariance matrix is adaptively adjusted in accordance with the observation residuals to further improve the robustness of the filtering algorithm. And by combining the continuousdiscrete model with the correction factor, the filtering accuracy and the antiabnormal measurement value are unified. The simulation results show that compared with the traditional robust algorithm, MARSRCDCKF can track targets more accurately and the robustness is more strong under conditions of singlepoint measurement abnormality and multipoint measurement abnormality.

Abstract:Aiming at the problem of missed detection and wrong detection in the process of SSD(Single Shot Multibox Detector)multiscale object detection, this paper proposes a object detection algorithm that incorporates multidimensional dilated convolution and multiscale feature fusion. Among the multiscale features output by the convolutional neural network, the shallow layer has more detailed information, and the deep layer has more semantic information. According to this feature, this paper adopt three types of multidimensional dilated convolution shallow layers for the shallow network. The feature enhancement module obtains a feature map with semantic information, downsamples the enhanced feature map, and fusion features of different layers; at the same time, a channel attention module is introduced in the deep network to assign weights to channels, suppress useless information, and improve object The detection performance. The research results show that the detection accuracy of the algorithm in this paper is 79.7% on the PASCAL VOC dataset, which is 2.4% higher than the SSD algorithm; the detection accuracy on the KITTI dataset is 68.5%, which is 5.1% higher than the SSD algorithm, and the detection speed reaches realtime requirements have effectively improved the missed and wrongly detected object.

Abstract:Aimed at the problems that the supervised learning model for image segmentation is long in training time, a large number of training samples is needed to ensure model accuracy requirement, and the labeling takes a lot of work, time and energy, an unsupervised image segmentation method is proposed based on neural network in different color space. Firstly, the images are transformed into different color space models to obtain the color representation of images in different color gamut spaces, and then by using felz and quickshift methods, a coarsegrained cluster is made after the images being transformed to form superpixel results and label each pixel accordingly. Finally, the finegrained image feature discrimination ability of neural networks is utilized for making a fineturning, obtaining the final image segmentation results. The method is validated on publicly available datasets such as COD10K selected, and the experiments prove that the proposed method can segment images reasonably with less inference time consumption and faster speed in comparison with the supervised training for a long time.