Research On Image Location Of Intelligent Mobile Target Based On Transfer Learning

National defense and army building are the security guarantee of our national economic development.Shooting training is very important to improve the combat capability of soldiers.Intelligent mobile targets can simulate realistic battlefield scenarios and help to improve the actual combat ability of trained officers and soldiers The core of intelligent mobile target design lies in the ability of autonomous positioning of moving target in space.The inaccurate positioning may lead to the failure of moving target to appear in the designated shooting area,and even to the potential danger of shooting safety in serious cases.In view of this,this paper studies and implements an intelligent moving target image classification and localization method based on convolutional neural network transfer learning with supervised data enhancement.The main research work of this paper is as follows:1.Three machine learning methods are used to classify and locate scenes in static space.By adjusting the size of the image and using color histogram equalization,and then flattening processing into one-dimensional row pixels,the sample set is preprocessed as feature vector output,and the classification and location results of three machine learning are obtained through experiments.A static spatial location method based on deep learning is proposed.The structure model and network parameters of convolution neural network are designed.The classification and location results of convolution neural network are obtained through experiments2.Aiming at the spatial classification and location of small sample data sets in complex dynamic space,a transfer learning method of convolutional neural network based on supervised data enhancement is proposed.The model of convolution neural network is Inception-V3 model,which has been trained,and the localization effect of single image localization and combined image localization are studied respectively.The experimental results show that the generalization ability of the network model can be enhanced by using the preset data transformation rule to expand and enhance the image on the basis of the existing data.The effect of model training and position prediction using the enhanced combined image data is the best.The accuracy of the four test sets of clear data,occlusion data,blurred data and variable light data can reach 99.67%,91.67%,95.67%and 98.83%respectively.The average prediction accuracy of the test set is 95.96%.3.Intelligent mobile target system is designed and built,which includes image positioning system and information management terminal system.It can carry out data communication,image acquisition,migration learning model training,spatial positioning,map coordinate display,mobile target information acquisition and management and other functions.Through the debugging and verification of the system's actual overall performance,it shows that the actual average prediction accuracy of the system on six indoor locations can reach 98.67%.