Study On Soldier Command Identification In Infantry Fighting Vehicle Environment

Speech recognition is an early research field of artificial intelligence.After years of development,speech recognition has achieved practical results in ordinary natural scenes.However,in the case of strong noise,due to the presence of background noise,the endpoint detection of the recognition system is invalid,and the recognition effect is completely unavailable.With the improvement of the intelligence of military training,it is of great significance to collect the data of soldiers’ training for the study.In the process of infantry fighting vehicle training,soldiers will give corresponding commands according to the training requirements,observed battlefield conditions and operational actions.The content and time of the commands are important indicators to evaluate the training level of soldiers.However,in the environment of infantry fighting vehicles,the noise of engine and track is very high,and the signal-to-noise ratio of speech signal is very low,so the existing password recognition algorithm cannot be used.In this paper,the recognition of soldiers’ operation commands in live ammunition training scenes of infantry fighting vehicles is studied.The specific work is as follows:1.A speech enhancement algorithm based on Increase Decrease Encoder Decode Full Convolution Neural Network(IDEDFCNN)is proposed.Due to the existence of strong background noise in the background of infantry fighting vehicles,the accuracy of password recognition is not only affected,but also the clarity of background monitoring of command post is reduced.Therefore,before password recognition,the soldier’s password data needs to be enhanced.To this end,this paper puts forward a lift decoding the convolutional Neural Network(IDEDFCNN)speech enhancement algorithm,in this algorithm,the input voice signal is preprocessed by framing and windaging,and the Fourier transform and logarithmic operation are used to obtain the Fourier logarithmic amplitude spectrum characteristics,and eight consecutive frames of voice signal input to the IDEDFCNN model.The model firstly uses the encoder to model the speech signals of adjacent frames,so as to obtain the information between the context of the signal,and then uses the decoder to establish the connection between the context information and the speech frames to be enhanced,so as to achieve the purpose of speech enhancement.Experiments show that this algorithm can achieve better speech enhancement effect.2.A password recognition algorithm based on Compress VGG16(CVGG16)network is proposed.Infantry fighting vehicles under the background of speech password recognition in essence is a kind of classification problem,in order to meet the requirements of warrior password recognition real-time and further enhance the accuracy in speech password recognition,experimental proof VGG16 cannot meet the requirements,the direct use of traditional,therefore,this paper presents a compression VGG16 network password recognition algorithm,the algorithm will increase after the soldiers password data of spectra unified to a reduced to the size of 112 * 112 images as input of CVGG16 network,The Block5 of traditional VGG16 is deleted,a full connection layer is removed,the depth and activation function of convolution kernel of each convolution layer are modified,and the task of warrior password recognition is completed by using the idea of image classification.The accuracy of password recognition based on CVGG16 can reach more than 90%.