Research On Target Behavior Cognition Technology Based On Machine Learning

As the combat environment becomes more and more complex and the combat are limited by time and space,the combat system should have efficient information processing speed and fast comprehensive response time.This requires that combat equipment should have the ability to accurately and quickly identify unknown targets,and the ability to further understand the intention of the target behavior from the information.Target behavior cognition is the basis of battlefield situation assessment and threat assessment.It has extremely important significance for improving combat capability and has become an important research topic at home and abroad.Target behavior cognition is a process from perception data to cognitive behavior.In previous researches on target behavior cognition,feature extraction was often manual,which was not only time-consuming and labor-intensive,but also unable to effectively process the massive information of the information system.Machine learning has strong self-learning ability and massive information processing ability.The knowledge graph technology with it as the core has also developed in full swing and achieved good results in many fields.Considering that target behavior cognition should be accurate and fast,it can be achieved with the help of machine learning’s powerful self-learning ability and massive information processing ability.Therefore,this paper combines machine learning methods to study target recognition and target behavior analysis.Aiming at the problem that conventional machine learning algorithms can not learn the laws of time series data,this paper studies the target recognition performance of LSTM networks with long-term memory capabilities,designs the LSTM network recognition model,details the construction and training of the model,and studies the noise sensitivity of the LSTM network,the simulation results show that the LSTM network has a high recognition rate under low signal-to-noise ratio and good noise robustness.In view of the fact that the actual acquired sensor data often has problems of varying degrees of missing,this paper proposes a mobile sliding window data filling algorithm based on principal component analysis.The principal component analysis is used to calculate the attribute value of the adjacent attributes on the missing items.In the sliding window data filling results,this not only considers the time series relationship of the data,but also considers the mutual relationship between adjacent attributes.The simulation results show that the filling algorithm has a good data recovery rate,and the LSTM network that can learn key information also has good recognition performance faced with the absence of data.Aiming at the problem that conventional information systems cannot effectively analyze,organize and use massive information,based on target recognition,this paper proposes a technology framework for target behavior mining and retrieval analysis based on knowledge graph,and obtains the target in the form of knowledge retrieval behavioral information.The framework organizes knowledge from the time dimension,considers the connection of knowledge between different time periods,combines the static attribute information of the target with the dynamic knowledge information,and mines and analyzes the target behavior based on the extracted knowledge information to achieve effective organization and full use of knowledge.Research is carried out using technical routes such as data collection,knowledge extraction,knowledge storage and retrieval,and knowledge graph analysis.In terms of knowledge extraction,entity recognition,relationship recognition,and attribute recognition are carried out based on domain words and auxiliary words combined with regular expressions to ensure the accuracy and robustness of knowledge extraction.Finally,the analysis of the formed knowledge graph and the retrieved information proves that the architecture has excellent performance.