Research On Multi-sensor Data Fusion Method Based On DS Evidence Theory

Multi-sensor data fusion technology is widely used in military and civilian applications because of its better fault tolerance,complementarity and high reliability compared to single sensors.As a branch of multi-sensor data fusion technology,Dempster-Shafer(DS)evidence theory has received wide attention from experts and scholars for its outstanding advantages in the representation and processing of uncertain information.However,as the research progresses,there are still some problems in applying DS evidence theory for multi-sensor data fusion,such as uncertain information modelling,the measurement of conflicts and the fusion of conflicting basic probability assignment(BPA or evidence)functions,real-time and sequential fusion of time-domain BPA functions,etc.These problems constrain the development of DS evidence theory,so this thesis uses the DS evidence theory framework as the basis,combines interval number theory,uses belief Hellinger distance,belief entropy and credibility decay model as technical means,designs a BPA function generation strategy based on interval number distance and model reliability,improves the weighted average strategy for fusion of conflicting BPA functions and the time-domain BPA function The adaptive processing strategy of credibility decay in fusion is improved to achieve the purpose of fusion of multi-sensor data air domain and time domain scenes based on DS evidence theory.The main contributions of this thesis are as follows.A BPA function construction method based on interval number distance and model reliability is proposed for the problem of how to transform realistic objective data into a BPA function in the DS evidence theoretical framework.Firstly,the idea of interval number is introduced to construct the interval number model of the attributes using the attribute data;secondly,the interval number distance is introduced to calculate the interval number distance between the attribute data and the attribute model,and it is transformed into the initial basic probability assignment(IBPA)function;finally,the IBPA function is modified from both the static and dynamic reliability perspectives of the model to obtain the final BPA function.Simulation results show that the BPA function construction method proposed in this thesis makes full use of the information contained in the original data and can effectively characterise the information with higher accuracy and anti-interference capability.In response to the problem that conflict coefficients cannot effectively measure the degree of conflict between BPA functions and that the fusion of conflicting BPA functions appears counter-intuitive,a weighted conflicting BPA function fusion method based on belief Hellinger distance and belief entropy is proposed.Firstly,the belief Hellinger distance is combined with a probability transformation function to measure the degree of conflict between BPA functions;secondly,a new belief entropy is defined to quantify the information volume of BPA functions based on the Deng entropy,taking into account the weight of multisubset propositions and the scale of the discriminative framework;next,the degree of conflict and information volume of BPA functions are combined to construct weighting coefficients to weight BPA functions averaging;Finally,fusion is performed by Dempster combination rule.Simulation results show that the proposed method in this thesis can effectively measure the degree of conflict between BPA functions as well as solve the counter-intuitive problem in the fusion of conflicting BPA functions.Compared with other methods,the proposed method has higher convergence speed and higher fusion belief,which is more conducive to subsequent decision-making.A time-domain BPA function fusion method based on an adaptive processing strategy is proposed to address the problem of poor anti-interference capability of the credibility decay model in time-domain BPA function fusion.Firstly,make full use of the cumulative BPA function information from the historical moment to predict the BPA function information for the current moment.Secondly,the difference between the predicted BPA function and the true BPA function is judged by the correlation coefficient and the true BPA function is discriminated against distortion.next,the real-time reliability of the information of the BPA function at the historical moment and the BPA function at the current moment is reasonably allocated by the adaptive processing strategy in combination with the credibility decay model.then,the historical moment BPA function and the current moment BPA function information are then discounted by real-time reliability.finally,the sequential fusion of time-domain BPA functions is performed using the Dempster combination rule.Simulation results show that the method proposed in this thesis has better robustness and anti-interference capability,and the fusion process satisfies the real-time and sequential fusion characteristics of the time-domain BPA function.