| In the field of modern air defense operations,the difficulty of evaluating and analyzing the threat level and combat intent of targets have increased with the increasingly complicated airfight environment and the broaden number and types of aerial targets,which affecting the decision-making of the command and control system.Intelligent recognition and trajectory prediction of aerial targets serve as an important basis for command and control systems to assess threat levels and analyze combat intentions.Effective identification of aerial target types and accurate prediction of future three-dimensional spatial positions are beneficial for improving the accuracy of command and control systems in assessing target threat levels and combat intentions.Overall,this research focus on the research of the key technologies in intelligent recognition and prediction of aerial targets,concentrating on the trajectory data of aerial targets.The main innovative points and work can be described asthe following aspects:1.A trajectory preprocessing method is proposed to address the problem of outlier points in trajectory data,uneven trajectory shape,and excessive burrs.This method is composed of a outlier determination and a correction algorithm.The correction algorithm designed by sliding average with variable sliding windows and a trajectory smoothing algorithm based on genetic algorithm improved cubic B-spline curve.Firstly,the Shoveler criterion is used to determine outlier points in the trajectory data,and the variable sliding window moving average method is used to correct outlier points;Secondly,the modified trajectory data is smoothed using a cubic B-spline curve improved by genetic algorithm.The experimental results show that this algorithm can effectively eliminate outliers,correct trajectory data and achieve trajectory smoothing.Besides,the propose also provides a good data foundation for subsequent recognition and short-term trajectory prediction.2.A dynamic feature based on intelligent recognition algorithm for aerial targets is proposed to address the problem of low accuracy in aerial target recognition under limited track data and insufficient feature extraction.Firstly,based on in-depth analysis of the motion characteristics of aerial targets,dynamic pressure mean features that can reflect the target’s dynamic performance are extracted,.In the meanwhile,a dynamic feature dataset is constructed by combining them with the height and velocity mean features in the preprocessed trajectory data.Secondly,a lightweight BP neural network is constructed as a classifier for intelligent judgment and recognition of aerial targets.The experiment results proved that the introduction of dynamic features improves the average accuracy of intelligent recognition of aerial targets by 18.125%.And the dynamic features are suitable for different classifiers,which shows good interpretability and representation ability.The accuracy rate for recognizing multiple types of targets is above 85%,with an average accuracy rate is 90%.3.Aiming at the problem of insufficient consideration of the correlation relationship between data in existing trajectory prediction methods and the cumulative error generated by recursive prediction strategies,which leads to low accuracy of trajectory prediction,a short-term air target trajectory prediction algorithm based on residual correction CNNBi LSTM hybrid neural network is proposed.Firstly,the algorithm innovates a convolutional module to extract potentially correlated spatial position features from target trajectory data,utilizes a bidirectional long short time memory network to extract temporal features from trajectory data,and achieves real-time single step prediction and multi-step advance prediction of aerial targets.Secondly,the integrated moving average autoregression is introduced as a residual correction model to modeling the residual generated by real-time single step prediction and calculate the residual of the hybrid neural network model for multi-step advance prediction.Finally,the output results of the hybrid neural network model and the residual correction model are fused to obtain the final trajectory prediction value.The experiment results confirm that this algorithm greatly reduces the errors generated by network models and recursive strategy predictions,and can also significantly improve the accuracy of short-term prediction of aerial target trajectories.4.A software for intelligent recognition and prediction of aerial targets based on trajectory data.This system integrates the main algorithm achievements in trajectory preprocessing,intelligent recognition and prediction of aerial targets in this research.Besides,this system can achieve multiple functions such as reading historical and real-time data,outlier correction,trajectory smoothing,dynamic feature extraction,intelligent recognition of aerial targets and short-term prediction of trajectories.According to the experiment results,the system can effectively identify the types of aerial targets and accurately predict the three-dimensional spatial position of aerial targets. |
