Research On The Method Of UAV Formation To Recognize And Position Sea Surface Targets

As an emerging combat platform,with the gradual development and maturity of autonomous navigation and control technologies,UAVs are playing more and more roles in many military fields such as reconnaissance,surveillance,communication relay,electronic countermeasures,fire guidance,combat outcome evaluation,and early warning.The more important the role.UAVs have the characteristics of small mass,small size,and strong overload capacity in mission execution,and have outstanding advantages compared with manned aircraft.In addition,the use of unmanned aerial vehicles to perform reconnaissance and surveillance tasks has the characteristics of being more flexible and not limited by geographic location.Therefore,it is of great significance to carry out research on target recognition and positioning methods based on UAV formations for efficient and safe completion of reconnaissance and surveillance tasks.This paper takes the UAV combat formation as the research object,and takes the accuracy of target detection and recognition and the positioning accuracy as performance indicators to carry out the research of autonomous target reconnaissance and positioning.It is based on neural network technology,information fusion technology and autonomous detection and recognition system design.The nonlinear multi-model filtering method is used to design the target positioning and tracking system,and the target detection,recognition and positioning modules are tested and evaluated in the task environment.First,systematically investigate the domestic and foreign research and development status of target detection,recognition problems and nonlinear filtering methods in UAV remote sensing images,analyze the current research status of target recognition and location problems,and summarize its task background,technical path and significance.Secondly,the problem of target detection in UAV remote sensing images is studied,and the basic structure of convolutional neural network target detector and the classic network YOLO are studied.Considering the problem of small target size,low resolution,low signal-to-noise ratio and sample imbalance among different types of data sets in UAV remote sensing images,a targeted loss function improvement method is given.In addition,in order to effectively simulate the combat environment and reduce the cost of data collection,a digital simulation environment based on a modeling rendering engine was built.The simulation experiment results show that compared with the original network,the improved network structure has higher accuracy in the case of low resolution and small targets,and can meet the task requirements.Then,the problem of low-resolution target recognition and UAV formation joint recognition in UAV remote sensing images is studied,and the structural characteristics of two classic neural network classifiers are analyzed and corresponding improvements are made for low-resolution situations.Considering that the information obtained by a single drone is limited,and the accuracy of the result cannot be further improved due to the limitation of the amount of information,the single-machine classification feature vector is compressed by the principal component analysis method,using this as the data source,and the data fusion classification based on recurrent neural network is given.It also completed the migration learning and performance evaluation in the public data set and the self-built data set of this article,which proved that the neural network classifier proposed in this article has sufficient accuracy in both the single drone and the drone cluster.Finally,based on the in-depth study of the interactive multi-model structure of the volumetric Kalman filter algorithm,an adaptive improvement method of variable dimensional parameters is proposed on this basis.Aiming at sea surface motion where a single model cannot change the motion pattern rapidly,and the volume Kalman filter has good approximation accuracy for nonlinearity,a recursive equation of interactive multimodel algorithm based on volume Kalman is given,and the filter is built to perform Mathematical simulation proves the effectiveness of the interactive multi-model algorithm for locating maneuvering targets on the sea surface.On this basis,the variable-dimensional parameter adaptive multi-model algorithm is studied to further improve the positioning accuracy.Simulation experiments show that compared with the traditional IMM method,the above-mentioned applied algorithm can effectively locate maneuvering targets on the sea surface and improve the positioning accuracy.