In order to solve the problem that the ADS-B ground station monitoring for the flight situation of the general-aviation aircraft makes it difficult for the aircraft itself to perceive the flight situation of the surrounding aircraft, a situation monitoring terminal system based on ADS-B IN is designed for general-aviation aircrafts. The system adopts a microcontroller to parse the received messages of ADS-B, and uses Android Studio to design terminal display software. At the same time, the improved Residual Mean Interaction Multi-Model（RMIMM） algorithm is used to perform track filter processing for the position and state information of the aircraft, thus to improve the flight state estimation accuracy. The improved algorithms and systems are validated by using the data received through actual devices. The results show that, the terminal system can effectively sense the flight situation of the aircrafts around the general-aviation aircraft and improve its flight safety.

Considering the problem of parameter uncertainty existed in hypersonic vehicles, a double-power sliding mode controller based on backstepping control is designed. Firstly, the dynamics model of the hypersonic vehicle is translated to the strict feedback form, and the dynamic surface control method is adopted in backstepping design to avoid the problem of explosion of differentiation terms. Secondly, the sliding mode control method is integrated on the basis of the backstepping control, and double-power approaching law is used in the sliding mode control to eliminate the chattering. The stability of the method is proved by the Lyapunov stability principle. The simulation results show that the double-power sliding mode controller based on backstepping control can realize stable tracking of altitude and velocity in the presence of uncertain parameters.

Aiming at the problem of real-time autonomous obstacle avoidance of underactuated USVs under unknown environment with multiple dynamics obstacles, an online obstacle avoidance path planning algorithm was proposed. Based on the guidance of the global path, the constraints of multiple obstacles, of USV ontology kinematics and dynamics, and of the established air route were taken into full consideration. At the same time, the multi-objective function was taken as the optimization function to perform online real-time obstacle avoidance through obstacle motion prediction and rolling optimization. The correctness of the algorithm was verified by simulation experiments and analysis. A certain USV was used to test the obstacle avoidance algorithm on the lake, and the test results verify the effectiveness of the algorithm in practical engineering applications.

Aiming at the formation keeping problem of Unmanned Aerial Vehicles (UAVs) in three-dimensional space formation flight after implementation of rendezvousing, a three-dimensional formation keeping system based on a virtual leader is proposed. Based on the three-dimensional kinematics relationship of UAV flight, and combined with the simplified UAV autopilot model, a relative kinematic model is established based on the rotating coordinate system with the virtual leader as reference. The UAV formation keeping controller is designed by the sliding mode variable structure control method. The simulation verification shows that, the controller can enable the UAVs to track the maneuvering virtual leader and maintain the formation stability, and the response accuracy and speed are ideal.

Aiming at the recognition of radar jamming emitters, we proposed a method for identifying the individual interference source by using the difference between each emitter's phase noise characteristics. By using information entropy and fractal theory, 39-dimensional features were extracted by multi-domain joint feature extraction method on the basis of time domain, frequency domain, wavelet domain and bispectrum domain. Considering that the direct use of classifiers for processing high-dimensional data is prone to fall into the “curse of dimensionality”, and in order to effectively remove the redundant information and maintain a high recognition rate, we proposed a feature dimension reduction method based on the fusion of ReliefF salgorithm, Fisher criterion and correlation coefficient criterion by use of the Sequential Forward Search (SFS) strategy. The optimal feature subsets of the corresponding original data were obtained. The experimental results show that, compared with the traditional feature selection method, the proposed method can effectively reduce the feature dimension and maintain high classification accuracy.

For attitude tracking of quadrotor Unmanned Aerial Vehicle (UAV) considering parameter uncertainty and external disturbances, a desired compensation, Robust Integral of the Sign of Error (RISE) control method with adaptive gain adjustment was developed. Firstly, the original cascaded model of quadrotor UAV attitude was derived in a strict feedback form with lumped disturbances. Secondly, the desired compensation RISE controller was designed for the quadrotor model described in strict feedback form. Finally, an adaptive law was introduced to on-line tune the sign function gain in the desired compensation RISE controller. The effectiveness and superiority of the proposed method has been validated through extensive simulations and comparisons.

Aiming at the problem that it is difficult to model and control the complex systems, this paper adopts a data-driven control strategy, uses the input and output data to establish a dynamic linearization model of the controlled system, and uses the sliding mode predictive control method to control the dynamic linearized model. In the method, the parameter estimation error of the system dynamic linearization process is regarded as generalized disturbance, and the strong robustness of the sliding mode predictive control reduces the influence of generalized disturbance on the system control, which effectively improves the control effect. An analysis is made to the stability of the control algorithm, and the stability and robustness of the system with external disturbances and time-varying parameters are verified by simulation. The results show that, compared with the existing algorithms of model-free adaptive control, model-free predictive control and model-free integral terminal sliding mode control, the proposed algorithm has better control quality.

In order to solve the trajectory tracking control problem of the quadrotor aircrafts, an Interconnection and Damping Assignment Passivity Based Controller (IDA-PBC) was designed to achieve the trajectory tracking control.Firstly, the mathematical model of quadrotor aircraft was developed by Newton-Euler method, and then the trajectory tracking task was implemented by introducing the contraction theorem to the IDA-PBC controller.Finally, the effectiveness, tracking performance and robustness of the proposed IDA-PBC controller were validated by computer simulations.The results showed that: 1) The IDA-PBC can complete the tracking control task of the quadrotor aircraft effectively;and 2) Compared with the PID controller, the IDA-PBC has the advantages of quicker convergence speed, lower tracking error and stronger robustness.With the IDA-PBC, the system has great steady-state performance and dynamic performance, which can meet the trajectory tracking control requirements of quadrotor aircrafts.

Most of the existing hyperspectral anomaly detection algorithms only use the spectral dimension information of hyperspectral images, which does not reflect the advantages of “image-spectrum integration” of the hyperspectral data, and may result in degraded detection performance.This paper proposes a hyperspectral anomaly detection algorithm based on non-local self-similarity（NLSSAD）.The algorithm first establishes dual stereoscopic windows, in which the inner window represents the 3D spatial-spectral structure window of the Spectral Vector of the Pixel to be Tested (SVPT).Then the stereoscopic window in the background is found, which is the most similar to the inner window, and the distance between the two windows is calculated to obtain the non-local self-similarity index of SVPT.The anomaly detection results show that, compared with the existing algorithms, NLSSAD has better performance in detection rate and operation speed.

Aiming at the problems of spectral distortion and blurring because of ignoring the image edge information, a Discrete Wavelet Transform (DWT) algorithm based on gradient sharpening is proposed.The IHS transform is used to extract the luminance component from the multi-spectral image, and then DWT is executed on the luminance component and the panchromatic image to acquire the high-frequency and low-frequency subbands.For the low-frequency subband, the gradient sharpening criterion is used according to the edge feature to implement the fusion of low-frequency coefficients.For the high-frequency subband, the way of maximizing the MSE of the pixel region is used to complete the fusion of the high-frequency coefficients.The results are obtained by inverse transformation of DWT and IHS.It is found through experiments that the spectral features and visual effects of the fused remote sensing image are better than that of the current spectral sharpening method.

For a class of linear systems with norm bounded uncertain parameters, the robust H∞ controller and state observer are designed in the case of actuator failures.Considering that the state of the system cannot be measured directly, the state observer is used to obtain state information unknown to the controlled system.Assuming that the actuator of the controlled system is failed, the design of the fault-tolerant controller and the state observer is discussed.According to the definition of robust H∞ controller and Lyapunov stability theory, the matrix inequalities satisfying the robust H∞ controller and state feedback controller are obtained.By using Schur's complement lemma, the matrix inequality is transformed into a standard linear matrix inequality.This linear matrix inequality can be easily solved by the LMI toolbox in Matlab.Finally, the effectiveness of the method is demonstrated by simulation examples.

The target detection based on deep learning technology extracts and learns multi-level features of the target through artificial neural network, and sends the feature map to the classifier to predict the category and position of the target.According to the model training methods, the algorithms can be divided into two types:one-stage detection algorithm and two-stage detection algorithm.In this paper, the representative algorithms of each stage are introduced in detail, and comparison and analysis are made to the algorithms on the PASCAL VOC data set.Finally, the development trend of target detection is forecasted.

The basic principle of light-guided Head-Up Display(HUD) technology is introduced with its technical advantages and feasibility.The current developing state of the technology at home and abroad is summarized.Then, an analysis is made to the critical technologies of the aviation light-guided head-up display, and the detailed indexes and parameters of the existed light-guided head-up displays are given through some examples.Finally, the technical bottleneck and potential application prospect are described based on the developing trends of the light-guided head-up display technology.

The accuracy of the traditional text classification method is not high enough for naval text classification task.According to the regularity of location distribution of key information in the naval military text, the traditional one-dimensional Convolutional Neural Network (CNN) is improved, and the naval text classification model is designed.In the aspect of one-dimensional convolution, a variable-step convolution method is proposed.The text features are mined by using low step-size at the beginning and end of the text, and high step-size in the middle, so as to improve the mining ability of key features at the beginning and end of the text.In the aspect of one-dimensional pooling, a weighted pooling method is proposed to convert text location information into weighted values to participate in the pooling operation, reflecting the importance of text location information.Experimental result shows that, compared with the traditional support vector machine, K-neighbour algorithm, the traditional one-dimensional CNN and the long/short-term memory network, our text classification model has higher accuracy rate,recall rate and F1 value.

With the increase of radio transmission rate, the visible images and the infrared images acquired by UAVs can be downloaded to the ground station at the same time, and more battlefield information can be obtained based on the fusion of the two kinds of images.The existing technology relies on the method of digital image processing, which requires a large computation cost and high performance of hardware.To solve this problem, the system uses two DM642 as the main CPU to process the infrared images and the visible images respectively.Firstly, the target features of the designated area in the infrared image are extracted, and then the geographic coordinates of the target area in the infrared image are calculated by using the attitude information from the UAV and the airborne camera, and converted into the screen coordinates of the visible image.Then, the target is corrected according to the installation angle and field-of-view angle of the visible-light camera and the infrared camera, and the infrared target is fused into the visible image according to the method of maximum pixel gray level.The result of flight tests show that, the system has fast image fusion speed, accurate target fusion position, which meets the requirements to the system.

Airborne sensors play an important role for the flight control system in the processes of flight status acquiring and control law solving of internal and external loops.In addition to increasing the hardware redundancy to improve the system reliability, it can also increase the analytical redundancy of the model to improve the system fault tolerance.In this paper, the flight control system is modeled and analyzed based on the aerodynamic parameters of UAV, and an optimal fault state estimator is designed.An improved residual detection method is proposed on the basis of Kalman-Bussy filter, which is used to online learn and estimate the fault state of the system.Furthermore, under the condition of joint voting based on periodic time and residual value, an adaptive reference model regulation law is designed to compare and modify the analytical model in real time, so as to reduce the influence of uncertainties caused by system interference.The simulation results show that the proposed method is feasible and effective.

The influence of radar tracking error on the miss distance of the command guided system is studied.Firstly, the radar tracking error is modeled and analyzed.The command guidance loop model with introducing of radar tracking error is established, and the analytical solution of the simplified model is obtained, according to which the relationship between the time and the miss distance caused by radar tracking error is analyzed.Secondly, the complex instruction guidance loop model is studied in detail by using the adjoint method, and the change of miss distance caused by the random error of radar tracking error with the time when the parameter value of each link in the guidance system changes is obtained through simulation.The result shows that:the miss distance of the command guided system caused by radar tracking error increases with the increase of guidance time; and the systematic error is the main factor affecting the miss distance within limits.

Aiming at the high subjectivity of traditional training evaluation methods for shore-to-ship missile weapon systems, a training effect evaluation index system of shore-to-ship missile weapon system is constructed based on network analysis.An “equivalent treatment” module is designed to reduce computational complexity, which realizes the mapping from weapon system training data to the secondary indicators.The Analytical Network Process (ANP) and Variable Precision Rough Set (VPRS) are used to establish the training effect evaluation model, and the expert mean confidence method is introduced to reduce the subjectivity and contingency of the evaluation process.The SD software is used to solve the model, and the weight coefficient of the training evaluation index of the shore-to-ship missile weapon system is obtained.The results verify the effectiveness of the method.

As a new type of neural network, Extreme Learning Machine (ELM) has the advantages of fast learning speed and good generalization ability.However, due to the random generation of its input weight matrix and hidden layer bias, it is easy to cause network instability and large deviation.An Improved Differential Evolution (IDE) multi-objective optimization algorithm is introduced to obtain the input weight matrix and hidden layer bias of the ELM with the smallest training error, so as to improve the performance of it.Finally, the optimized extreme learning machine is applied to fault diagnosis of a rudder system.The results show that the improved ELM has high learning speed and diagnostic accuracy.

Focusing on the ship target detection of remote-sensing images, a real-time remote sensing detection method based on YOLO-v3 was proposed.By introducing the spatial pyramid pooling structure, and by using the dense connection and Inception structure to achieve dimension reduction transition module, the extraction of network feature information was enhanced.By re-replacing the connection structure of the backbone network and optimizing the multi-scale feature fusion detection, a new network structure was designed, which reduced the parameter quantity, enhanced the feature transmission, and finally achieved an effect better than the original method.The remote sensing image dataset provided by Airbus in the Kaggle competition was used for the comparison test.The result showed that, the average detection accuracy of the algorithm is 84%, the accuracy is improved by about 4% compared with the original algorithm, and the speed of detection reaches 23 frames per second.

In computer vision research, traditional small field-of-view, visible-light cameras have limited access to information and poor adaptability to special environments.In order to increase the amount of acquired information and improve the adapatability to the environment, it is necessary to use a binocular ultra-wide-angle infrared camera.Aiming at the problem of low calibration accuracy of binocular ultra-wide-angle infrared camera, based on a general camera model, and by use of the active infrared radiation calibration plate, a calibration method is proposed.Through the calibration in the camera model, the internal parameters and external parameters of the binocular ultra-wide-angle infrared camera are accurately calibrated.The experimental results show that the focal length, baseline distance calibration error and the mean reprojection error of the calibration corners are lower than 0.99%, 0.24% and 0.60 pixels, respectively.The high calibration accuracy verifies the correctness of the proposed method.

In order to satisfy the demand of flight simulation training and establish a realistic visual simulation environment, it is necessary to make image mosaic and fusion to multi-channel projection.Most of the existing technologies depend on the hardware support of the projector itself.In order to lower the requirement to the hardware environment, we studied the existing mosaic algorithms and made some improvements.A more reasonable and efficient method was used to accomplish curved-surface generation and projection transformation to the images produced by scene rendering module, thus to make the image projected onto the spherical screen through the projector be a normal image.Then, transition processing was made from bright to dark to the brightness of overlapped part of the images from several projectors, to make sure of the smooth transition of images of different projectors.The verification result showed that, the image processed by the mosaicing and fusion system can achieve a seamless display effect of multi-channel projection on the spherical screen, which achieves the expected qualification.