Research On Two Types Of Data Fusion Algorithms And The Applications In Target Tracking

As an important part of data mining,data fusion combines the characteristics of various aspects of data to extract the greater value of data,so it has been widely used in computer interaction,health monitoring,target tracking and other fields.A data fusion algorithm based on robust fuzzy C-means clustering(RFCM)is a classic and traditional fusion algorithm which has the advantages of automatically classifying sample data and fast clustering by applying RFCM to data fusion.At the same time,the data fusion algorithm based on deep long and short time memory network(DLSTM)is a fusion algorithm based on deep learning which has the advantages of maintaining the relationship between data and handling the long-term timing series well.But any single fusion algorithm has limitations.Therefore,this paper innovates and improves the deficiencies of RFCM and DLSTM algorithms,and the improved algorithms are proposed.The main work of this paper is as follows:1.A data fusion algorithm based on adaptive fuzzy C-means clustering is proposed.For the RFCM algorithm,there are some inadequacies such as inaccurate fusion and low reliability of integration.This paper introduces the adaptive coefficient,combining the Kalman filter principle with the neural network prediction method based on multi-layer perceptron,the error covariance is estimated,and a fusion strategy is designed.A data fusion algorithm based on adaptive fuzzy C-means clustering(ARFCM)is proposed.Through the simulation experiments of synthetic datasets and real datasets with different properties,the ARFCM algorithm is compared with FCM,RFCM,RLS,KF,EFCM,FDPFCM,and IIFCM algorithms which verifies the advantages of the ARFCM algorithm in fusion error and fusion efficiency.2.A hybrid multi-sensor data fusion algorithm based on fuzzy C-means clustering is proposed.For the AFCM algorithm,there are some problems such as randomly initialize the cluster center and easy to fall into the local optimal.In this paper,a similarity measure method for arbitrary data sets is given.The local density metric of Gaussian kernel function is used in CH-CCFDAC algorithm to determine the initial clustering center.A fusion strategy is designed based on AFCM algorithm.A hybrid multi-sensor data fusion algorithm based on fuzzy C-means clustering(HFCM)is proposed.The comparison simulation experiment with the CH-CCFDAC,RLS,KF,AFCM,FCM,ARFCM fusion algorithm in the synthetic datasets and UCI database shows that the HFCM algorithm has obvious advantages.3.An eye tracking data fusion algorithm based on long and short time memory networks is proposed.In view of the insufficiency of DLSTM algorithm for parallel computing and the inability to obtain global information,this paper performs feature extraction and feature processing based on the characteristics of eye movement and tracking data.By introducing CNN into DLSTM network,a new network structure is developed and a fusion strategy is designed.An eye tracking data fusion algorithm based on long and short time memory network(Eye LSTM)is proposed.By comparing the fusion effect of Eye LSTM algorithm with two deep learning algorithms MLP and DLSTM on 10 sets of real eye movement data sets and 10 sets of real tracking data sets,the experimental results show that the Eye LSTM algorithm performs good in terms of fusion quality.Overall,the algorithms proposed in this paper have achieved good results,but there are problems such as single data features,multiple parameters,and the efficiency of the algorithm needs to be improved which are the next step to be studied.