一种基于Resnet的非稳态数据关联方法

Data correlation algorithm based on Resnet for unsteady data

设计了一种基于残差网络的非稳态数据关联方法.利用残差网络对多层次和异构特征的提取能力求解数据关联问题中的最优决策函数.首先将全局关联问题分解为固定问题空间大小的基本关联问题,然后设计深度网络提取非稳态数据中的不变特征,找到基本关联问题解的分类模型,可在误差分布变化和参数无法准确估计的情况下,提高关联鲁棒性和准确性.仿真试验结果表明:当误差分布参数在一定范围内变化和未知条件下,本算法优于联合概率数据关联(JPDA)算法和K近邻(KNN)算法.

A data correlation algorithm based on Resnet for unsteady data was designed in this paper.Resnet was used to extract multi-level and heterogeneous features to solve the optimal decision function in correlation algorithm problem.First,the global correlation problem was decomposed into the basic correlation problem with fixed problem space size.Then,the deep network was designed to extract the invariant features of the unsteady data and find the classification model of the solution of the basic correlation problem.In this way,the correlation robustness and accuracy can be improved when the error distribution changes and the parameters cannot be accurately estimated.The simulation results show that the algorithm designed in this paper is effective.When the error distribution changes within a certain range and under unknown conditions,it is better than joint probabilistic data association(JPDA)and K-nearest neighbor(KNN)algorithm.