基于Deeplabv3网络的飞机目标微动信号分离

Aircraft Target Micro-Motion Signal Separation Based on Deeplabv3 Network

针对空中飞机目标的微多普勒效应提出一种基于时频图与深度神经网络分离直升机、螺旋桨和喷气式三类飞机旋转部件和机身的方法。本文从飞机目标时频图像素差异着手,根据深度学习语义分割网络提取飞机目标时频掩膜图,将掩膜图与飞机目标多分量时频矩阵进行乘法拟合,实现三类飞机目标多分量信号分离。通过建立的仿真数据集进行多组实验,结果表明对飞机目标多分量信号,深度学习语义分割网络提取时频掩膜的方法能够很好地分离机身和旋转部件信号,并起到抑制杂波的效果。

Aiming at the micro-Doppler effect of airborne aircraft targets,a method based on time-frequency diagram and deep neural network to separate rotating parts and airframes of helicopters,propellers and jets is proposed.This paper starts from the difference of the pixel value of the aircraft target time-frequency diagram,extracts the aircraft target time-frequency mask map according to the deep learning semantic segmentation network,and multiplies the mask map with the multi-component time-frequency matrix of the aircraft target to realize the multi-component signal separation of the three types of aircraft targets.Through the established simulation data set,multiple sets of experiments are carried out.The results show that the method of extracting the time-frequency mask for the aircraft target multi-component signal and the deep learning semantic segmentation network can separate the signals of the fuselage and rotating components.Through the established simulation data set,multiple sets of experiments are carried out.The results show that the method of extracting the time-frequency mask for the aircraft target multi-component signal and the deep learning semantic segmentation network can separate the signals of the fuselage and rotating components,and has the effect of suppressing clutter.