The radar can invert the parameters of sea evaporator ducts by using the sea clutter information.In order to improve the performance of evaporative duct inversion, an improved Particle Swarm Optimization (PSO) algorithm is proposed.When taking the minimum value, the objective function established between the actually measured radar sea clutter power and the clutter power calculated by the evaporation duct model can reverse the profile parameters closest to the actually measured evaporation ducts.According to this idea, the inertia weights and learning factors are adaptively adjusted based on the basic PSO algorithm, and the adaptive compression factor is introduced to ensure the fast convergence of the algorithm, thus the high-precision evaporation duct parameters can be obtained.Simulation experiment proves that:compared with basic PSO algorithm, the improved PSO algorithm has better global convergence performance, and the inversion speed is obviously improved when dealing with large-scale data.

Time-Triggered Ethernet (TTE) sends communication tasks at regular time triggered by a reasonable scheduling timetable to ensure the real-time and safety-critical performance.This paper proposes a scheduling algorithm of TTE based on time-slot partition.The algorithm obtains information of the switches in the network topology, and partitions the scheduling timetable of each end system.According to the routing information of TT traffic, the sending time is configured in the corresponding partition, so as to ensure the conflict-free transmission of TT traffic in the physical link.At the same time, the scattered free time-slots are combined to reduce the delay of RC traffic and to avoid the waste of network resources.Experimental results show that the proposed algorithm has advantages in network resource integration and computing speed while ensuring the schedulability of TT traffic.

The consistency control problem of isomorphic multi-agent system is considered in most of the research on multi-agent systems, but in practical application, a variety of agents need to work together according to the desired formation shape, and the dynamics of real systems often have nonlinear properties.Aiming at this problem, a heterogeneous nonlinear multi-agent system time-varying formation control strategy based on PI controller and adaptive control strategy is designed by using nonlinear parameter decomposition method under undirected topology.Since the real system will inevitably be affected by external disturbances, and there is a certain range of external disturbances.Therefore, on the basis of the designed control strategy, the stability of the multi-agent system with external bounded disturbances is considered.The stability of the designed control strategy is analyzed by constructing a Lyapunov function and verified by making simulation on Matlab.The results show that the time-varying formation control of heterogeneous multi-agent system can be realized by the designed control strategy and it has certain effectiveness.

From the perspective of the complex demand for commanding and decision-making in modern electronic warfare, a quantitative aided decision-making model is established based on the dynamic inference ability of Dynamic Bayesian Network (DBN), and the one-dimensional quantitative analysis in traditional operation decision-making is transformed into quantitative, dynamic multi-dimensional analysis.The factors that have effect on the operation effectiveness are taken for building the BN at each operational moment.The dynamic inference of the auxiliary decision-making is realized by using HMM model inference algorithm based on the conditional probability matrix of the nodes.Simulation and analysis have been made to the actual exercise examples, and some practical and effective recommendations are obtained for commanding and decision-making of electronic warfare commanders, which demonstrates the effectiveness of the model.

Aiming at the problems of slow convergence speed and easy divergence of state variables in the trajectory tracking of a quadrotor aircraft, a dual-loop hybrid finite time control strategy is proposed.According to the Newton-Euler equation, the dynamic model of the quadrotor is deduced, and is then divided into two control loops, i.e., the inner and the outer loop according to the principle of time scale.The outer loop uses a finite time control strategy to accelerate the convergence speed of the three-axis position and the yaw angle.The inner loop uses fast nonsingular terminal sliding mode control technology to achieve rapid convergence of the attitude angle.A virtual prototype of the quadrotor aircraft is built to simulate its trajectory tracking control effect in a three-dimensional simulation environment.The simulation results show that, compared with the other two common controllers, the controller designed in this paper has the best control accuracy, robustness and tracking effect, and it can better meet the requirements of quadrotor aircraft trajectory tracking control.

Aiming at the problem of trajectory planning exploration when the starting point and ending point are known and the environmental information is unknown, an improved sparse A* UAV trajectory planning exploration method incorporating pheromone is proposed.Based on the local map information obtained by lidar and by introducing the pheromone, the cost function in the sparse A* algorithm is optimized to realize autonomous avoidance of obstacles in unknown environment without repeated exploration.On this basis, a physical implementation approach based on Robot Operating System (ROS) is proposed.Under spiral-form obstacle scene, it is verified that the method can avoid repeated exploration of the environment.In addition, the method is extended to the columnar obstacle scene.The experimental results show that the method can effectively achieve the exploration of trajectory planning in the unknown environment.

Towards the establishment of effectiveness evaluation indicator system for complex equipment, the one-dimensional convolution neural network is introduced, and a parallel processing framework with multi-scale convolution kernels is constructed.Then, the original data of equipment is learned adaptively from multiple dimensions, then the characteristics are integrated into a new effectiveness evaluation indicator system, which avoids limitation of subjective factors and computational difficulty, and lays a foundation for intelligent methods of effectiveness evaluation.

Considering that the N-step phase shift method cannot overcome the nonlinear response of the projection equipment in the measurement system, and in order to suppress the jumping error of the three-frequency heterodyne phase unwrapping, we proposed a 3D measurement method based on double-N-step phase shift and full-frequency phase unwrapping.Firstly, the initial and fused wrapped phases are solved by standard double-4-step phase shift method, and the combined wrapped phases are determined according to the fringe modulation system.Then the first-order fringe series is obtained by using the three-frequency heterodyne method, and the unwrapped phase of the second-order fringe is obtained based on the relationship between fringe series and phase in the full-frequency phase-unwrapping algorithm.The experimental results show that the proposed method is simple in implementation, smooth in phase expansion and has no jumping error.Compared with the existing methods, the proposed method has better phase unwrapping effect.

A task risk assessment method combining variable weight theory with cloud model is proposed.Aiming at the deviation of the evaluation results caused by the extreme values that may appear in the task, the interval variable weight function is constructed, and the normalization standard of risk factor data is established.Aiming at the excessively strong influence of Ex in traditional cloud models on the assessment results in risk assessment, the interval form is adopted to weaken it, and the method for determining the digital features of the interval cloud model is improved based on the attribute mathematical theory.The rationality and effectiveness of the method are demonstrated through examples.

A new nonsingular fixed-time sliding mode control method is proposed for the trajectory tracking control of the manipulator under the influence of such factors as unmodeled dynamics and random interferences.Firstly, the dynamic equation of the n degree-of-freedom manipulator is derived by using Kanes equation and virtual work principle.Secondly, a new fixed-time sliding mode surface is designed based on the fixed-time stability theory.Combined with 6R manipulators Kane dynamics model, a non-singular fixed-time sliding mode controller is designed aiming at the uncertainty of model parameters and external interferences of the manipulator.The stability of the system is proved by using Lyapunov method.Numerical simulation shows that the designed controller can ensure that the convergence of the system state is independent of the initial conditions, and it has faster convergence rate, shorter convergence time and good robustness.

This paper focuses on time alignment algorithms in airborne multi-sensor data fusion, and theoretical analysis is made to the alignment error caused by non-synchronization of time.On this basis, in order to improve the accuracy of interpolation and extrapolation algorithm, an improved subsection linear interpolation algorithm is proposed to solve the time alignment problem, of which the data obtained from different sensors has big differences on time.Simulation analysis is made to the data fusion processing between airborne data link and fire-control radar in confrontation between two sides.The simulation results show that the proposed algorithm has fine applicability to the processing of airborne multi-sensor data fusion.

This paper aims to develop a robust optimal control method for missile autopilot systems suffering from external disturbances by using adaptive dynamic programming algorithm.First, a nonlinear disturbance observer is designed to estimate unknown external disturbance.Then, considering an integral sliding mode surface and the output of disturbance observer, an integral sliding mode controller is designed to make the system enter into sliding mode motion along the sliding surface.Subsequently, for equivalent sliding mode dynamic systems, a critic network is constructed with a novel weight update law, and the adaptive dynamic programming technique is employed to learn an optimal controller.With the Lyapunovs method, the stability of closed-loop system and the convergence of estimation weight for critic network are proved.Finally, the missile autopilot system is used to verify the effectiveness and feasibility of the designed approach.

Electronic warfare is the beginning of contemporary information-based warfare.The control of the battlefields multi-time, multi-space and multi-dimension information determines the outcome of the war.Decision-making issue in the information-based battlefield is different from the traditional combat command issue.Due to the complexity of the battlefield environment, the diversity of information and the information dependency of equipment, the research on interference decision-making is becoming more and more important.At present, the field of artificial intelligence is developing rapidly.The concepts of cognitive radio and cognitive electronic warfare have been proposed one after another.Based on the existing research results, this paper introduces the theory of cognitive interference decision-making, and sorts out relevant domestic and foreign literature.An analysis is made on three aspects of cognitive interference decision-making theory, namely, state recognition prediction, parameter control and resource scheduling.The commonly used theoretical methods and their existing shortcomings are summarized.Finally, the prospect of the radar cognitive interference decision-making is presented.

Aiming at the key problem of limited computing power of the system onboard UAVs, a trajectory planning method based on real-time octree map is proposed for indoor autonomous exploration.Firstly, the depth images obtained by the RGB-D camera are used as input to construct the octree map in real time.At the same time, the octree map in the current field of view of the UAV is scanned in real time for scene recognition.Based on the scene of autonomous recognition, a method for planning the next target point in different scenes is specifically designed, which effectively reduces the amount of airborne calculations, and realizes autonomous exploration of unknown indoor environments.The simulation experiment results show that the method has advantages on feasibility and computational complexity.

This paper studies the design of reduced-order observers for a class of linear discrete-time Markov jump systems with unknown inputs.First, a new reduced-order observer is designed for the linear system with unknown input, and the unknown input is completely decoupled by constructing the observer gain matrix.In addition, the sufficient conditions for the existence of the observer are given in the form of linear matrix inequality, and the observer error is proved to be finite time stable, which guarantees the good transient properties of the estimation.Finally, an example verifies the effectiveness and feasibility of the proposed method.

The variability of the aperture of the airborne opportunistic array radar enables its radio frequency radiation power control range be arbitrarily directed in space.This paper first presents the multi-aperture characteristics of an aircraft opportunistic array radar, and shows that the continuous changes of its multiple aperture values can be approximated by trigonometric functions.Then, in order to relax the strong constraints on the prediction error covariance during target tracking and improve target tracking performance, the ratio of Blackman window to Hamming window is designed as the relaxation factor of the prediction error covariance.Finally, the dwelling time optimization model of airborne opportunistic radar is analyzed and established based on the requirements of RF stealth.The simulation results show that the use of interactive multi-model and dwelling time optimization design based on the posterior Cramer-Rao bound constraints not only has better RF stealth performance, but also can save the resources and improve the target tracking performance of the airborne opportunistic radar.

The equipment Remaining Useful Lifetime (RUL) prediction method based on dropout Neural Network (dropout NN) has low precision since it uses a Bayesian Neural Network (BNN) with fixed priori distribution and approximate posterior distribution.To solve the problem, we proposed a RUL prediction method based on BNN with Gaussian approximation posterior distribution and a RUL prediction method based on mixed Gaussian-Bernoulli network.The former introduces the mixed Gaussian distribution as priori distribution and then optimizes BNN by unbiased Monte Carlo estimation of parameter gradient, while the latter introduces a discretized Gaussian prior distribution to define KL divergence correctly, and then optimizes the BNN.The verification results on PHM 2012 bearing dataset show that the mixed Gaussian-Gaussian network has better effect than dropout NN, which proves that BNN with the changed distribution combination can obtain better prediction effect.

It is difficult to obtain the infrared image of the aircraft target under air-to-ground background, which may leads to over-fitting or other problems of the algorithms for infrared aircraft target detection.To solve the problems, a data augmentation method for small-sample infrared aircraft targets is proposed based on the related methods of feature fitting and inter-channel attention mechanism of Generative Adversarial Network (GAN).Firstly, a special pyramidal multi-scale GAN structure is used to learn the feature information of a single image at different scales.Secondly, for the small infrared aircraft images, the structure of the network generator is improved, an inter-channel attention module is added to the generator to enhance the representation of small sensory fields and enrich the details of the generated images.Finally, the learning rate of the small-scale stage of the pyramid is scaled in the training phase of the network, to avoid the distortion of the generated images due to the overlarge learning rate.Several common target detection algorithms are simulated and evaluated, and the results validate the effectiveness and superiority of the proposed method by comparing it with the traditional data augmentation methods.

The “position+velocity” transfer alignment method is adopted to simulate the impact of two factors, airframe structure deformation and baseline information delay, on the real-time posture accuracy of the Inertial Measurement Unit (IMU) at the Antenna Phase Center (APC).The conversion relationship between main and sub inertial navigation coordinate systems is clarified, and the influence rules of these two types of factors on the real-time posture information of the sub-inertial navigation at the APC after transfer alignment are obtained.The simulation shows that:1) The matching algorithm is not suitable for the case where the airframe has intense dynamic deflection and deformation and with high frequency in deformation;and 2) The baseline information delay may seriously affect the estimation time and accuracy performance of the “position+velocity” transfer alignment.Therefore, it is necessary to make structural deformation compensation and delay compensation and use other measures when the above two factors have serious effects.

Low cost is an important attribute of UAV swarm.Aiming at the limitation of traditional UAV object locating technology, which requires that a laser rangefinder be equipped on the photoelectric platform, a joint realization method for target detecting and locating without relying on information from the laser rangefinder is proposed.Firstly, the object detection network based on deep learning is used to identify the object from UAV aerial images, so as to obtain the position information of the object in image coordinate system.Secondly, the coordinates of the object under aircraft coordinate system is calculated according to the relative height of the UAV, parameters of the photoelectric platform, and the position information of the object in the image. Then, the position information of the object in aircraft coordinate system is converted to longitude and latitude information under WGS-84 coordinate system.Finally, the validity of the proposed method is verified by actual flight of UAVs.The results of actual flight prove that the proposed method can effectively complete the object locating task without relying on the laser rangefinder, which provides a technical support for the low-cost UAV swarm.

Due to the absorption and scattering of underwater light, the underwater images often have color distortion, low contrast and fuzzy details.Therefore, a weighted fusion based multi-space transformation method is proposed for underwater image enhancement.Firstly, white balance processing is made to the image, which can effectively correct the blue (green) color appearance of the image.Secondly, the RGB space is converted into LAB space by the white balanced images, and the L channel is processed with the improved adaptive Gamma correction, and converted back to RGB space.Then, CLAHE and bilateral filtering are carried out in RGB space.After that, the RGB space is converted to HSV space, and the single scale Retinex algorithm combined with guided filtering is used to process the V channel, and then converted back to RGB space.Finally, weighted fusion and detail enhancement are carried out to the above three results, and the final enhanced image is obtained.Experimental results show that the proposed algorithm is effective for improving the contrast and sharpness of underwater images and removing color distortion.

Aiming at the problem of low recognition rate of single-kernel Support Vector Machine (SVM) for ship target classification in Synthetic Aperture Radar (SAR) image, a method for ship target recognition based on multi-feature extraction and Multi-Kernel Learning (MKL) SVM is proposed, which improves the accuracy of target recognition from two aspects of feature extraction and classifier training.First, the public data set is selected to extract the multiple types of features of the ship target, and then weighted fusion is made to the multi-kernel functions to construct a multi-core SVM model.Finally, multi-feature data are used to train and recognize the ship target.In view of the information redundancy problem in multiple sets of target features, the correlation coefficient is used to remove some highly redundant features and reduce the feature dimensions.The Particle Swarm Optimization (PSO) algorithm is used to solve the problem of kernel parameter selecting of the SVM kernel function.The experimental results show that the proposed method effectively improves performance of ship target recognition, and the comprehensive recognition rate is increased from 87.18% of the traditional SVM to 92.31%.