Research On Unmanned Drone Cluster Target Deduction Technology Based On Knowledge Graph

As an important part of artificial intelligence,knowledge graph describes concepts,entities and relationships among entities in the real world in the form of symbols,thus providing the ability to analyze and solve problems from the perspective of relationships,so that knowledge can be accessed(search),queried(question and answer),and supported actions(decision-making).In addition,the deduction technology of the knowledge graph is one of the most core technologies.By deducing the unknown knowledge in the knowledge graph,the depth and breadth of the knowledge graph can be expanded,the problems of incomplete and incomplete knowledge graph can be solved,and the automatic update and self-improvement of the knowledge graph can be realized.Therefore,the application of knowledge graph and inference technology based on knowledge graph to the target recognition application scenarios of UAV(Unmanned Aerial Vehicle)cluster can greatly improve the autonomy and intelligence level of UAV.In this paper,aiming at the problem of automatic construction of ontology and knowledge map in military equipment field,knowledge is extracted from a large amount of information through in-depth learning method to complete the construction of knowledge map.This paper puts forward the entity extraction model based on bidirectional long-term and short-term memory neural network combined with conditional random field and the neural network relationship extraction model based on attention mechanism.The system extraction model is implemented.The experimental results show that the accuracy of biLSTM-CRF model is 11.50%higher than that of CRF model,and the F1 value of biLSTM-Attention model is 8.21%higher than that of CNN model.In addition,in view of the shortcomings of the existing remote supervised relation extraction model,such as semantic loss and noise interference,this paper proposes a relation extraction model based on bidirectional gating cycle unit and multi-level attention mechanism,and introduces the composition of the model in detail.By predicting the implied relations in the knowledge graph,the knowledge graph can be perfected.Finally,the knowledge deduction system is designed and implemented,and the equipment knowledge map and remote supervision relationship extraction model are applied to the system.In addition,this paper proposes a multi-dimensional feature collaborative inference recognition algorithm based on knowledge map,which can realize the task of UAV cluster autonomous recognition by fusing multiple features to deduce the model of target equipment.Through the function test and performance test of the knowledge deduction system,it is proved that the knowledge map constructed in this paper is effective and the deduction recognition algorithm is efficient.