摘要
针对现有空中目标机动模式识别算法鲁棒性和抗噪性差的问题,提出了利用卷积神经网络直接对航迹数据进行非人工特征提取,从而实现机动模式识别的算法。针对目标机动段难以分割的现实情况,提出了滑动时间窗口的模式识别方法,并给出了基于滑动时间窗口的机动模式识别流程。对空中目标进行了航迹仿真,并进行了数据预处理,为卷积神经网络提供了合理训练样本。通过仿真实验确定了适合于机动模式识别的卷积神经网络的结构和参数,实验结果表明,构造好的卷积网络对机动模式的识别率达98.4%,并且在结合机动触发点后,对连续航迹的识别取得了良好效果。
In order to improve the air targets maneuvering pattern recognition accuracy,aiming at improving the system property of the robustness and resistance to noisy data, convolutional neural networks were applied to extract non-labor features from trace data directly for accomplishing the pattern recognition. Slide time window maneuvering pattern recognition method was proposed to solve the difficulty on maneuvering segmentation in actual situation, and then flow process diagram was given. Trace simulations were carried on after typical maneuvering pattern and parameters analyzed which provided reasonable training samples for convolutional neural networks. Suitable convolutional neural networks structure and parameters were identified after simulation experiments. The simulation results showed that the pattern recognition ratio was 98.4% on segmented pattern with the trained networks, at the same time, good results were achieved on continuous trace combined with the maneuvering trigger point.
出处
《微型机与应用》
2015年第22期50-52,56,共4页
Microcomputer & Its Applications
基金
国防"十二五"预研项目(40405070102)
关键词
机动模式识别
卷积神经网络
滑动时间窗口
航迹仿真
数据预处理
maneuvering pattern recognition
convolutional neural networks
slide time window
trace simulation
data preprocess
