Regarding the problem that when the area to be detected is large, the use of satellite or Unmanned Aerial Vehicle (UAV) alone will lead to the problems of low detection accuracy and slow detection speed respectively, a cooperative target detection method of “satellite-UAV” based on YOLO detection model is proposed.Firstly, the satellite equipped with YOLOv4-tiny model with channel pruning is used to quickly and preliminarily screen the target area in a large area.Secondly, the UAV equipped with YOLOv4 model with four-scale detection branches is mobilized to further accurately detect the areas containing the targets.The experimental analysis shows that the improved YOLOv4-tiny model can screen large-scale areas faster, and the improved YOLOv4 model has higher detection accuracy for the screened areas with targets.The cooperative detection of satellite and UAV can effectively combine the “fast” of the former with the “accurate” of the latter, and improve the detection efficiency.

Regarding the problem that the magnetic target single-point location method based on magnetic gradient tensor is greatly influenced by the estimation error of geomagnetic field, a magnetic target location method based on a cube array is proposed.This method uses the measurement information of magnetic gradient tensor of the upper and lower planes of the cube array to establish the equations about the target location, uses the characteristic of small geomagnetic gradient to eliminate the interference of geomagnetic field by making a difference in the magnetic field vector term, and then uses the difference of magnetic field vector and the information of magnetic gradient tensor to solve the position coordinates of the target.The simulation results show that the method can overcome the problem greatly influenced by the estimation error of geomagnetic field in existing methods, and improve the target locating accuracy.

The detection of small floating targets in strong sea clutter is a hot topic and difficult problem in maritime radar detection.Traditional algorithms have poor detection performance in a single domain (time domain, Doppler domain and time-frequency domain).Feature-based detection algorithm is an effective means to detect small floating targets on the sea surface.A feature fusion detection algorithm based on full polarization is proposed.Six features in time domain and Doppler domain are selected, six features based on full polarization are extracted, and the separability of features is analyzed.Finally, PCA-based anomaly detector is used for fast target detection, and is verified on the measured IPIX radar data set.The experimental results show that compared with the existing PCA-based algorithm, the proposed algorithm has better detection performance.

An Unmanned Aerial Vehicle (UAV) path planning method based on improved Rapidly-exploring Random Tree (RRT) algorithm combined with Artificial Potential Field (APF) and genetic algorithm is proposed to overcome the disadvantages of strong randomness of search range and slow convergence rate.Firstly, the target bias is introduced to guide the generation of random sampling points, and the target points have a certain probability of becoming sampling points, thus reducing the number of samples.Meanwhile, the APF is introduced to improve the generation direction of the new node, and the direction of the resultant force between the target point and obstacles is taken as the growth direction of the search tree, which improves the efficiency of path search.Then, a group of track points generated by the improved RRT algorithm are used as the initial population of the genetic algorithm, and a fitness function model is established.The genetic algorithm is applied to optimize the path, and the optimal path is obtained, which solves the problem of path randomness.Finally, the simulation results show that the path generated by the improved algorithm is shorter in length and consumes less time.

With the gradual maturity of the target detection algorithm based on deep learning, its deployment on UAV has become a hot topic.Regarding the problems of low detection accuracy caused by many small targets and easy occlusion in UAV aerial images, complex detection scenes and large scale variability, an improved algorithm based on the proposed S-YOLOv4 is introduced.Firstly, SENet attention mechanism is added on the original feature extraction network structure to improve the models ability to concentrate on useful information and enhance inter-channel attention.Secondly, a new detection layer with a resolution of 160×160 is added to refine the grid for better detection of small targets.Finally, the loss function is improved, and the class smoothing label is applied to the classification loss to reduce the penalty of negative samples and improve the generalization ability of the model.Compared with that of the original algorithm, the mAP of the proposed algorithm is improved by 3.4% under the real-time detection speed.

Regarding the problem of the Informed-RRT* algorithm of collision with dynamic obstacles in path planning, an improved path planning algorithm based on Informed-RRT* and artificial potential field method is proposed.The algorithm introduces elliptical region sampling strategy and adaptive step-size strategy to improve the stability and efficiency of finding feasible global path, and the optimal cost feasible path scheme is obtained in the boundary region of static obstacles.When the robot encounters dynamic obstacles in its movement according to the global path, the cutting path branching strategy and artificial potential field method are introduced to re-plan the local path to realize dynamic obstacle avoidance.The improved algorithm is applied to the simulation environment, and the results show that the improved algorithm achieves the functions of global optimal search and local obstacle avoidance, which verifies the effectiveness of the algorithm.

In order to realize the homogenization extraction of image features and improve the local aggregation of feature points, an improved AGAST feature extraction algorithm is proposed.By constructing the Gaussian pyramid of the image, the scale invariance of the feature points is realized.Then, the quadtree method is used to divide and screen the feature points, the homogenized distribution of the feature points is obtained, and the operation efficiency is improved by adapting the depth of the quadtree.The grayscale centroid method is used to calculate the direction of feature points to realize rotation invariance.The experimental results show that compared with other algorithms, the improved algorithm effectively improves the speed and uniformity of feature point extraction, and the extraction rate and homogenization are improved by 12.31% and 7.8% respectively.

Aiming at the disadvantages of traditional aerial target threat assessment methods, such as over-reliance on expert experience, large amount of prior knowledge needed, and unreasonable target attribute weights, an aerial target threat assessment method based on combined weighting method and TOPSIS method is proposed.For aerial threat targets, in order to comprehensively consider the subjective and objective weights of target attributes, a combined weighting method based on game theory is introduced to determine the integrated weights of target attributes, and then the TOPSIS method is used for aerial targets to conduct threat assessment. Finally, through the case simulation,the feasibility and effectiveness of the method are verified.

Regarding the problem of radiator threat assessment under incomplete information, an assessment algorithm based on IOWA-TOPSIS is proposed.Firstly, in view of the incomplete information of enemy radiator which is easy to occur in actual combat, the Induced Ordered Weighted Averaging (IOWA) null value estimation algorithm is introduced to predict the missing data based on the correlation of attribute sets.Then, Criteria Importance Through Intercriteria Correlation (CRITIC) is used to analyze the influence degree of different attributes on the assessment process and assign weights objectively.Lastly, the threat degree of different radiator is measured by Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) algorithm, and the threat level is ranked.The simulation results show that the algorithm can effectively evaluate the threat of radiator, and the results are more objective and accurate than those of traditional assessment methods.

A new model-free adaptive control algorithm is proposed by using a data-driven strategy to address the problem that it is difficult to control a nonlinear system with non-uniformly refreshed input and periodically sampled output.Firstly, the system output data is filtered by using a tracking-differentiator, and then the dynamic model is established at the equivalent dynamic equilibrium point by using this input-output data, and a model-free adaptive controller is designed.Secondly the penalty coefficients in the control law and Pseudo Jacobian Matrix(PJM)estimation are iteratively optimized by using the most rapid descent method, and the stability of the algorithm is analyzed.The advantages of it are:Reducing the influence of disturbance error and improving the ability of parameter optimization.Finally, the effectiveness of the proposed method is verified by an example of simulating a discrete-time nonlinear system with non-uniform sampling.

An algorithm based on non-local prior is proposed for the identification of large targets such as the distribution of enemy radar stations, the location of caverns and depots, and transportation supply line in mountainous areas.Firstly, the color index is used in the mountain haze image to obtain the haze line for cluster detection, and then the initial value of the transmission coefficient of pixels in the haze line is estimated.Finally, the transmission coefficient is regularized by the optimized algorithm, and finally the haze-free image is obtained.

Multi-source image fusion is widely used in various visual fields, which integrates the feature information of multiple images into one image.In the past decades, the methods of multi-source image fusion have been increasing, but they still cannot fully meet the needs of users.In order to design a fusion method with excellent performance in terms of accuracy, robustness and efficiency, it is necessary to comprehensively and systematically review and analyze the classic and the latest technologies so as to find new breakthroughs.Following the process of multi-source image fusion, the classification, framework and some methods of image fusion are introduced, and a new classification system of image fusion methods is proposed from a different perspective.Then, a comprehensive summary is made on the fusion evaluation methods.Finally, several applications of image fusion are presented, and the future development trend of multi-source image fusion is prospected.