Research On The Key Technology Of The Multi-sensor Information Fusion

After several decades of development,multi-sensor information fusion has become increasingly mature and widely used in the military field and civilian field,but there are still some pain points.First,this paper introduces the basic principles,the categories of information,functional models,fusion levels,processing structures,and key issues of multi-sensor information fusion.Second,for some key technologies,this paper conducts related research and is mainly divided into the following three aspects:First,in the direction-finding cross-positioning system,a single sensor cannot position the target independently,and it requires multiple sensors crossing to position the target.The process of crossing generates a large number of intersections,while the intersections contain a large number of false points.How to eliminate false points and associate data that come from the same target becomes a problem To solve this problem,this paper proposes an data association algorithm of using redundant angle by eliminating false point elimination.The innovation of this algorithm is mainly to construct an alternative set through the pre-association of two sensors to eliminate a large number of false points and to simplify the association set.Finally,the improved algorithm presented in this paper is compared with the data association algorithm based on residuals,the data association algorithm based on line-of-sight distance,and the data association algorithm based on redundant angle information.The simulation results show that the proposed improved algorithm is superior to the other three algorithms.Second,in multi-sensor information fusion,the process of sensor of detecting target is affected by noise or interference in the surrounding environment,resulting in the measurement data containing a certain amount of error and a certain percentage of outliers,which seriously affects the data quality.To solve this problem,this paper proposes an improved filtering algorithm based on the conventional Kalman filtering algorithm.It not only solves the problem of the outliers,preventing the filter divergence,but also introduces the observations of next time into the state estimation of the current time,revising the state estimate.Finally,the improved filtering algorithm proposed in this paper is compared with the conventional Kalman filtering algorithm.The simulation results show that the proposed algorithm has excellent performance.Third,the machine learning algorithms for target recognition is superior to traditional methods,but the construction of features is very difficult,especially in this paper,the scene is complex and the data is obscure.Extracting feature by manual work can make use of the expert's historical experience,prior knowledge,and professional background knowledge,and the features generally have good stability and effect,but it is time-consuming and laborious,and it easily leads to the omission of excellent features.The construction of features by model combine features on the basis of set of expert features to obtain set of more refined higher-order features,but ignore low-order features and artificial combinatorial features that cannot be learned.For the advantages and disadvantages of the set of expert features and the set of the combinatorial features,this paper fuses the set of expert features and the set of the combinatorial features to obtain a more comprehensive and more detailed fusion feature set.Finally,the author compares the set of fusion features with set of expert features and the set of the combinatorial features proposed in this paper,and the simulation results show that the proposed method has excellent results.