Research On Anticipated Recognition Technology Of Ground Attacking Target Based On Trajectory Prediction

At present,the problem of the command information system’s situation and lack of situation is prominent,and the technical products for situation analysis are seriously inadequate.It is difficult to quickly complete the understanding and analysis of complex situation information by manual alone.In response to the above situation,the use of intelligent technology to study the intention recognition method for the ground attack on the land battlefield,provide accurate and rapid decision support for the combat command,and seize the combat opportunity.Combat intention recognition is a process of reversely inferring the enemy’s combat intentions from changes in relevant situational elements.Traditional research uses expert knowledge to es Tablelish an empirical model.However,due to the influence of the commander’s personal decision-making preferences,combat units may exhibit different behaviors under the same combat intention.Therefore,it is difficult to es Tablelish explicit rules to accurately characterize the strong nonlinear influence brought by the personal factors of the commander,which leads to deviations in the intention recognition results.In response to this problem,using data-driven as the guiding ideology,neural network as the technical means,mining the inherent correlation characteristics from the multi-time enemy target state sequence data,on this basis,the samples are marked by experts,and the samples are matched with each other by fitting the samples.The mapping relationship of the tags completes the identification of target combat intentions.The main research contents are as follows:(1)Analysis of the process and difficulty of intent identification: starting from situation estimation,the research level of intent identification is analyzed;the mapping relationship between the commander’s combat intention and the change of relevant situation elements in the battlefield is clarified,and the process of intent identification is obtained.The key factors that affect the results of intention recognition are analyzed emphatically,and on this basis,the difficulties of ground target intention recognition are further studied.(2)Trajectory prediction technology for ground attack targets: A trajectory prediction model based on the Encoder-Decoder architecture is designed using neural networks.Collect data from the simulation system to build a data set,and process the category data in the data set through ONE-HOT coding technology;use Long Short Term Memory(LSTM)neural network to build a model encoder;use fully connected neural network and attention Mechanism to construct the decoder;finally,the high-dimensional feature tensor is mapped to the lowdimensional space through a fully connected neural network without activation function to output the predicted coordinates of the enemy target position at the next moment;finally,the performance of the model is analyzed based on the experimental results.(3)Intent recognition technology for ground-based targets: Using neural networks as a means,a Transformer-based intent recognition model is designed.The Encoder in the original Transformer is retained as the intention recognition model encoder,and the residual network structure and layer standardization technology are introduced on the basis of the multi-branch self-attention module to alleviate the over-fitting phenomenon of the deep network;the decoder is designed based on the fully connected network.The back-propagation algorithm trains the model;evaluates the performance of the model from the training loss curve,training time and test accuracy,and verifies that the model has good intention recognition accuracy.