Information Fusion Of Multi-measure From Single Station In Passive Ranging

In the military field, The passive positioning systems is good at concealment, which is attracting a growing body of reseach. Passive ranging can be divided into single station and multistation. Compared with the multistation, although the coverage area of the single observer is limited, it is not bound by other stations. There won't have a great effect on the performance of entire system by hitting one observation station. Therefore, mining and making full use of the single observation station information has an important practical value. Multi-measure is multidimensional measurement based on single platform, which obtains the target's measure information of multiple channels, and reflects the current state of target from different angles. The multi-measure passive ranging can obtain more comprehensive information. Therefore, we adopt information fusion method of multi-measure from single station for the tracking and location of the moving target.Based on the principle of partial coordinates, we process the target information of different motion models by using partial coordinates. Thus, we can obtain the information of bearings-only, pure elevation and distance-measured only. At the same time, we analyzed observability of the target's multi-measure and the theorems of observability of different motion models were obtained which provided the theories foundation for the target tracking under the condition of partial coordinates. When the target is observable, its multi-measure is simulated.Filtering is one of the basic elements of target tracking system. We filtered target information on the single model and the sudden change of speed or direction in actual movement by using impoved Particle filter and GM-PHD filter. The simulation experiment results of the two filtering algorithms show that on the tracking of bearings-only and pure elevation, GM-PHD algorithm is better than the improved resampling particle filter; however, on the tracking of distance-measured only, the improved particle filter is better than GM-PHD algorithm. At the same time, we verify this conclusion on the condition of different observation errors. Obviously, we can obtain accurate target position by fusing these two results by information fusion methods which can make full use of different measures of target information.Under the multi-measure of target's different motion models, we fused the information by the traditional BP neural network algorithm and PSOBP algorithm. The fusion errorof BP algorithm is relatively great because of its defects, such as slower convergence speed, falling easily into local optimal solution and so on. the run time of PSOBP algorithm is long. Because it would need much time to calculate the fitness of each particle to find the optimal individual and global optimal particle. In reaction to the phenomenon, we have proposed a new method by combining support degrees function matrix with BP neural network. The relative difference between the data is introduced into initial weights of BP neural network which improve the defect that BP algorithm is sensitive to the initial weights to a certain extent.Simulation results showed that: the fusion error of single-measure fused by the improved BP algorithm is small, the accuracy of the information fusion of multi-measure is higher. The BP neural network of optimization algorithm presented in this paper can track the target well in the situation when the measure dimension decreased. The run time of improved BP algorithm is slightly greater than BP algorithm's, but the location accuracy is far better than that.