The characteristics of global coupling and pulse synchronization of Pulse Coupled Neural Network (PCNN) can be used to solve the problem that the selection of high-frequency sub-band coefficients of fused images does not conform to human visual characteristics. However, PCNN has many problems in multi-source image fusion, such as complex model structure and complicated parameter setting. As to the model structure of PCNN, two kinds of methods for model optimization are analyzed, and the general law of PCNN when applied in multi-source image fusion is summarized, which provides a reference for better application of PCNN in multi-source image fusion.

The 3D micro-structures have important application value in the fields of optical trap, stealth and light field regulation, ultra-high pressure sealing and artificial biological materials.Advanced manufacturing methods for 3D micro-structures, including laser direct writing, photolithography, electronic and chemical etching, are rapidly developing. The detectors of Scanning Electron Microscopy (SEM), Atomic Force Microscopy (AFM), laser confocal microscopy, and low-coherence interference microscope are used, but these instruments are monopolized by foreign countries. This paper describes the research progress in low-coherence interference microscopy in China, including the optical design method of interference microscope, the assembly process, the wide-spectrum low-coherence interferogram processing and the micro-structure restoration method, and the evaluation of the uncertainty is given. The results show that the combined uncertainty of 3D morphology of the instrument is 0.9 nm and the maximum micro-structure axial resolution is 0.13 nm.

The convex optimization fusion graph theory is used for modeling analysis, the sensor array signal model is constructed, and the noise and interference subspace steering vector correction is designed.The Semidefinite Relaxation(SDR) and S-process convex optimization analysis are used to model the nonlinear non-convex problem as the Semidefinite Programming(SDP)convex optimization problem, and the covariance matrix model of spatial power spectrum matching is constructed.The robust wave formation method based on convex optimization about spatio-temporal data in the sensor network is obtained.The TDOA and FDOA polymorphic fusion algorithm is proposed to significantly improve the quality of multi-target tracking and achieve precise positioning.Through simulation experiments, it is shown that the proposed method has obvious advantages, and the resolution ability is superior to that of traditional positioning algorithms.

Due to the inherent speckle noise of SAR image and the influence of complicated terrain at the sea-land boundary, it is easy to incorrectly extract the coastline of SAR image by using the region merging method. In order to solve the problem of incorrect segmentation caused by a single scale, an algorithm for coastline segmentation of SAR image based on SLIC super-pixel is proposed.Super-pixel segmentation is used to preprocess the image at first.Then, the improved merging cost based on spectral and texture information (CT-Model) is used to merge, and finally the segmentation results are obtained. The coastline at the sea-land boundary is used for the comparison of segmentation effects. The experimental results show that the improved merging method is more accurate in coastline segmentation than the spectral-histogram method.

The stability loop of the Satcom On-the-Move (SOTM) servo system is strongly nonlinear due to the influence of backlash friction and other factors.Although the standard Back-Propagation (BP) algorithm has good identification effect for nonlinear systems, it has some shortcomings such as slow network convergence, process oscillation and poor generalization ability.To make up for the shortcomings, an improved adaptive BP algorithm based on the cumulative error function gradient and the double learning step is proposed to accelerate convergence and reduce oscillation.The global error threshold is also set to control training times in order to further enhance its generalization ability.Experiments are designed on the SOTM platform, and it is verified that the improved algorithm has obvious improvement in convergence, identification accuracy, the generalization ability and so on.A more accurate BP network model with nonlinear stability loop can be obtained.

Distributed Integrated Modular Avionics(DIMA)system has become the development direction of the next generation of avionics systems.In order to determine the network communication technology of the DIMA system, we built the system architecture model and discussed its difference and connection with the federated system and Integrated Modular Avionics (IMA) system on the Communication, Navigation and Identification (CNI) integrated system.On this basis, the network communication requirements of the DIMA system was summarized.The advantages of using Time-Triggered Ethernet (TTE) to realize DIMA system network interconnection in real-time and reliable communication as well as system incremental upgrade were verified by comparing and analyzing Fiber Channel (FC) and avionics full duplex switched ethernet with TTE.

Based on Model Predictive Control (MPC) algorithm, an active Fault Tolerant Control (FTC) method is proposed, which can diagnose sensor hard fault implicitly.Firstly, a prediction model library is established, which includes normal models and abnormal models with sensor faults.Then, the best matching prediction model is selected and the switch command is sent to the switching unit by the monitoring decision-making unit according to the sensor measurements, so as to realize online fault tolerance.At the same time, active fault-tolerant control simulations of CMPASS-40K turbofan engine are carried out.The results show that the monitoring and decision-making mechanism can judge sensor faults and realize fault-tolerance in a relatively short time.Moreover, this method can not only realize the tolerance of known faults, but also realize the tolerance of unknown faults.

The coherent blanking jamming technology is widely applied in the penetration jamming equipment, against which the radars of missile defense systems have few effective countermeasures.Based on existing radar technology, the paper proposes a new method for the countermeasure of coherent blanking jamming on the basis of radar Auto Gain Control (AGC) by studying the inherent defects of the coherent blanking jamming technology based on intermittent sampling.The principle analysis and simulation show that the method is effective in the confrontation of coherent blanking jamming.

In view of the noise and interference of the ViBe algorithm in the detection of long-distance moving targets on the sea surface, we propose four improvements based on the ViBe algorithm.First, the range of sample values in background modeling is expanded to reduce wrong judgments.Then, different thresholds and background model update rates are adaptively set according to the degree of dynamic change of the background, and the adaptive ability of the algorithm is enhanced.Next, a blinking level is defined.When the blinking level reaches a certain threshold, it is determined to be a blinking point, which reduces the misjudgment of the blinking point on the sea surface.Finally, pyramid-transformation is made to the input video sequence to obtain images with different resolutions.The improved Vibe algorithm is used to detect the foreground at different resolutions, and the detection results are fused to reduce the influence of blinking points on the sea surface.The experimental results show that the proposed algorithm can reduce noise on the sea surface and detect and extract long-distance moving targets on the sea surface with good real-time stability.

The accuracy of the attitude information of the quadrotor directly affects the flight control effect of the aircraft.This paper introduces the attitude calculation principle of the fusion algorithm combining gradient descent with complementary filtering and the adaptive complementary filtering algorithm commonly used in quadrotor aircraft.The static and dynamic filtering effects of the two algorithms are measured and compared in an independently-developed IMU system.The experimental results show that the adaptive complementary filtering algorithm is superior to the fusion algorithm of gradient descent and complementary filtering on filtering effect in both cases.

To improve the speed of Staple target tracking algorithm, dimension reduction of the extracted image features is carried out by using Principal Component Analysis(PCA)in the process of learning translation filters. In the process of learning scale filters, the number of extracted samples with different scales is reduced from 33 to 17, and scale reduction is carried out by using QR-factorization. Scale response number is increased to 33 by interpolation to ensure scale estimation accuracy. Simulation results show that detection accuracy is almost unchanged and the tracking speed is improved by about 50% in comparison with Staple algorithm.

The air-to-ground multi-target firepower assignment is the core of the airborne multi-target attack system.The characteristics of air-to-ground multi-target firepower assignment problem are analyzed, and different optimal evaluation criteria are proposed.According to the corresponding objective functions, three air-to-ground multi-target firepower assignment models are established.Based on the adaptive ant colony algorithm, the genetic operator is used to carry out cross-mutation operation, and the pheromone updating rule is modified by referring to the idea of Particle Swarm Optimization (PSO) algorithm.The steps of the fusion algorithm to solve the air-to-ground multi-target firepower assignment problem are given.The effectiveness of the fusion algorithm is verified by simulation experiments.

Due to the development of stealth technology and the requirement of attack tactics, the carrier aircraft may enter the enemy missiles inescapable attack area.To solve the problem of maneuvering evasion of the aircraft in the inescapable attack area of radar-type medium/long-range air-to-air missiles, a model of evasion decision-making based on deep-learning network is designed.A maneuverable flight database is established by using Monte Carlo simulation method, and the network model is learned.The effectiveness of network output is verified by simulation.The simulation results show that, the network output after learning can satisfy the condition of maneuverable flight and has better decision-making effect in the pursuit-evasion simulation, which can significantly improve the probability of successful escape in the missiles inescapable attack area and provide a reference for the pilots maneuvering decision-making when they perform high-risk missions.

For the quad-rotor UAV to achieve accurate landing on the slope in the complex environment, a novel system framework is proposed.The quad-rotor UAV with Manifold and visual sensor guidance is used to realize smooth slope landing.The autonomous landing control algorithm of the drone is optimized by using the Proportional-Integral-Derivative (PID) control method.After Manifold calculates the navigation data in real time, the quad-rotor UAV can accurately reach the position above the landing point.It is proposed to screen the data collected by the quad-rotor UAV platform DJI M100 through the cost function, and accurate landing on the slope is accomplished by using the reverse compensation algorithm.The experimental results show that the proposed algorithm can effectively maintain the stability of the quad-rotor UAV during flight and can accurately land on the slope.

This paper studies underwater image processing in closed space based on dark channel prior and adaptive color correction. For the problems of poor image processing effect and color distortion of traditional underwater image dehazing methods in closed space with uneven illumination intensity, this paper introduces adaptive red channel compensation parameters to avoid the phenomenon of red light loss after dehazing the underwater image by using dark channel principle, and uses the vignetting model to improve the brightness of underwater image. The channel is calibrated to reduce the influence of light intensity on image quality. The proposed method is tested in a real sealed underwater environment. The experimental results show that, compared with traditional dehazing algorithms, the proposed method can effectively improve the brightness of the image while restoring the color of the image.

Aiming at the problem that the traditional ant colony algorithm has slow convergence speed and may fall into local optimum in the path planning process, the ant colony algorithm is combined with the artificial potential field method added with virtual traction and fast function, and the resultant force of the potential field is introduced as part of the heuristic information of the ant search path point.The combined algorithm has higher global search ability, avoids the local optimum problem caused by the misleading of the heuristic information in the traditional ant colony algorithm, and improves the convergence speed.In order to verify the effectiveness of this method, a simulation experiment with Matlab software was carried out.The result shows that the robots motion trajectory is smooth and is close to the optimal path.

A ship-based Electro-Optical Tracking System (EOTS) driven by PMSM through gear system is designed to reduce the effect of ship-swaying caused by winds and waves.The fiber optic gyroscope is used to detect the motion state of EOTS in real time, and the EOTS is stabilized relative to inertial space by servo control unit.Unlike the land-based EOTS, the swaying motion of a ship caused by winds and waves is coupled to the ship-based EOTS, which has a great impact on its tracking performance.A Disturbance Observer (DOB) of the ship-based EOTS is designed to reduce the effect caused by ship swaying, and of the disturbances of backlash, dead zone, friction, system model mismatch, etc.The DOB is introduced to estimate the lumped disturbances of the system, and the estimated value is used in the feed-forward compensation design to improve the disturbance rejection ability.The experimental results show that the proposed scheme can ensure good anti-disturbance ability, which can enhance the control performance of the system.

Based on related standards of CAN bus, the airborne bus standard ARINC825 makes adaptive improvements on CAN protocol from the aspects of communication scheduling, integrity protocol and redundancy, which meets the requirement of civil aircrafts on high safety.To ensure determinism, major and minor time slices based communication scheduling are introduced to the high layer protocol.The determinism of ARINC825 is analyzed, a design method based on FPGA is put forward, and the design is then verified through network simulation and tests, which demonstrate that the design and the end-to-end latency meet the requirements of determinism.

In the simulated flight test of the missile control system, the test efficiency is low and the flight condition of missile can not be simulated.To solve the problems, a closed-loop flight simulation test scheme is proposed, which greatly satisfies the requirements for flight test of the rapidity, full coverage and effectiveness for flight test of the missile control system, and can test the coordination between the different parts of the system and the ability of the entire system to adjust to large disturbances.This paper introduces the necessity, principle and implementation method of closed-loop test, and discusses the design of the excitation signal algorithm and the method of judging missile performance based on test data.The rationality of this test scheme is verified by simulation.

In order to restore a motion blurred image to a clear image, a blind image deblurring method based on dark channel and low-rank priors was proposed.Firstly, the sparsity of dark channel was used to estimate the intermediate restored image, and weighted low-rank prior constraint was introduced to suppress the noise in the intermediate restored image, so as to improve the accuracy of the fuzzy kernel estimation.Then, the accurate blurred kernel was obtained by the optimization strategy of alternating iterations.Finally, the Hyper-Laplacian prior method was used to get the clear image.The experimental result shows that the proposed method can effectively suppress noise and ringing effect, retain image details, and has a good restoration effect for a motion blurred image.

An efficient high-quality soft shadow rendering method based on adaptive light source sampling is proposed.First, multiple shadow maps are generated by regular sampling of some areas of the surface light source, and then the shadow maps are re-projected to the viewpoint.The shadow difference value in image space is used to determine if it is necessary to further subdivide light source sampling points, so as to realize adaptive layer-by-layer sampling and avoid generating shadow maps that do not contribute to soft shadow quality.The proposed algorithm can not only keep the soft shadow realistic, but also improve the rendering efficiency.

Single-photon detection technology uses a single photon as the information carrier, which breaks the existing laser detection limit, and is currently a hot topic in applied basic research at home and abroad.When using InGaAs/InP APD (Avalanche Photodiode) to detect 1064 nm laser, there is a large dark-counting rate and after-pulse probability, which affects the accuracy of detection.Compative analysis is made to the effects of three quenching methods on the performance of single-photon detection circuit.Gated quenching has less dead time than passive quenching and active quenching, and has better inhibition of dark-counting rate and after-pulse probability.For the gated quenching method, four schemes of sinusoidal gating of self-differencing, dual-APD balanced detection and capacitance balance are studied to effectively reduce the spike noise generated by the gating signal.By optimizing and debugging the detection circuit of the sinusoidal gating method, the dead time of the detection circuit is 9.3 ns.When the detection efficiency is 9%, the dark-counting rate is 1.64×10-6 /ns and the after-pulse probability is 3%.