针对传统的航迹融合算法精度较低、计算过程需要先验状态估计的缺点,提出了一种基于卷积神经网络(CNN)的航迹融合算法。各局部航迹在融合中心已经过时空校准和航迹关联。由于目标运动轨迹具有时间相关性的特点,采用连续多周期的局部航迹估计,结合深度学习积累经验的能力,解析出当前时刻的更精确的系统航迹估计,实现航迹融合。实验表明,该种融合算法能够处理具有共同过程噪声复杂环境干扰下的综合误差,并且在不同传感器和环境情况下,以相同的CNN模型结构训练,融合后的系统航迹误差均方差都低于各局部航迹误差均方差,证明了该算法能够提高航迹精度,具有可行性。

Due to the low accuracy of traditional track fusion algorithm and the need of prior information in the calculation process, a track fusion algorithm based on convolutional neural network (convolutional neural network, CNN) is proposed. Each local track has been time-pace calibrated and track associated at the fusion center. Since the track of target is time-dependent is adopted, combined with the ability of deep learning to accumulate experience, continuous multi-eriod local track estimation to calculate more accurate current track estimation, and achieve track fusion. Simulation results confirmed that this fusion algorithm can deal with the comprehensive error under the interference of complex environment with common process noise. Furthermore, under different sensor and environment conditions, with the same CNN model structure training, the mean square error of the track fusion error is lower than that of each local track’s which proves that the algorithm can improve the track accuracy, and is well robust.

V2789.3