Behavior Recognition Of Group Communication Based On Deep Belief Network Algorithm

Along with the development and application of modern information technology,the battlefield environment has gone into more informative and diversified.In modern battlefield operations,the communication information among fighters formations,fleets formations and group objects interior is mostly transmitted by radio.In the battlefield intention recognition,intercepting communication information of combat groups to distinguish behaviors intentions plays a vital role to make battle plans and war decisions quickly.With regard to the communication behaviors problem of combat groups,the simulation of groups communication behaviors is carried out.The external features data of the different communication behaviors is calculated.The communication behaviors data set is obtained.The behaviors recognition model is established with Deep Belief Network(DBN).The recognition accuracy of groups communication behaviors is calculated.This paper main contents are as follows.1.The data set of communication behaviors is obtained by simulating the communication behaviors of combat groups.For fighters formations,fleets formations and combat UAV formations,the communication relationships and configurations under different behaviors are analyzed.The external features of the communication behaviors are obtained.The communication behaviors script is written to formulate the communication behaviors rules.According to the behaviors rules,the behaviors simulation interface is set to calculate the behaviors features data.The communication behaviors database is obtained.The data preprocessing is exploited to quantify data.A description method of groups communication behaviors is offered.The communication behaviors data set of different combat objects in different networking modes is obtained.2.The parameters of Deep Belief Network are optimized.The nodes numbers construction method,and different encoding and decoding structures are adopted to optimize the parameters of Deep Belief Network.The implementation of Restricted Boltzmann Machine(RBM)and Contrast Divergence(CD)algorithm are summarized.The encoding and decoding structures of RBM network are analyzed to compare the features learning processes of RBM network.The construction of DBN model and the pre-training of network are analyzed.The construction method is adopted to combine the number of the hidden layers nodes.The network depth selection problem is studied by combining reconstruction errors and test accuracies.The better algorithm efficiency and recognition effect are obtained through processing network parameters and coding structure.3.The simulation experiment and recognition result analysis of the communication behaviors are finished by DBN algorithm with optimized parameters.The feasibility of DBN algorithm is verified by the standard data set.The identifiability of the behaviors is analyzed by the distribution characteristics of the behaviors data.The feasibility of DBN algorithm is validated by comparing the recognition effect of support vector machine based on genetic algorithm and Bayesian network.The DBN models are trained with different network parameters and coding struetures to compare the reconstruction errors and behaviors recognition rates.The validity of the optimized parameters DBN algorithm is validated by simulation experiments and the analysis of recognition results,and the recognition rate is higher.The recognition rate of the groups communication behaviors is 95.75%.