常规空中目标轨迹识别方法能够识别单一轨迹样式,但对于复杂轨迹样式识别效果差。文中提出一种基于Softmax多分类网络的空中目标自适应分段轨迹识别算法。首先,通过滑窗在时间上遍历整条复杂轨迹序列,在窗内利用设计的拟合方程对窗内轨迹进行拟合,得到方程参数;然后,基于方程参数训练Softmax分类网络模型,完成基于方程参数的窗内轨迹样式识别;最后,考虑到窗内识别错误及其特征,将识别结果进行同类合并,并基于Haar小波重构对识别结果进行修正,消除离散错误识别结果。仿真结果表明:文中所提方法能有效地对复杂轨迹样式进行分段识别。

Conventional aerial target trajectory recognition methods can recognize single trajectory pattern, but the recognition performance for complex trajectory patterns is poor. In this paper, an adaptive segmental trajectory recognition algorithm for aerial targets based on Softmax multi-classification network is proposed. Firstly, the sliding window traverses the whole complex sequence in time, and the designed fitting equation is used to fit the trajectory in the window to obtain the equation parameters. Then, the Softmax classification network model is trained based on equation parameters to complete the pattern recognition of trajectory in the window. Finally, considering the recognition errors in the window and their characteristics, the recognition results are merged in the same category, and the recognition results are modified based on Haar wavelet reconstruction to eliminate the discrete error. Simulation results show that the proposed method can effectively segment the complex trajectory patterns.

TN957.52