基于压缩感知和深度学习的分类识别技术

Classifying Aircraft Based on Compressive Sensing and Deep Learning

雷达对空飞机目标分类可实现雷达装备获得敌机属性和类别信息,对于现代战争其重要性显得尤为突出。针对复杂电磁环境下的飞机目标分类问题,结合防空雷达的特点建立3类(固定翼、螺旋桨和直升机)飞机旋转部件调制回波模型,并理论分析了不同类型飞机目标的微动特征差异。仿真分析在复杂电磁环境下干扰对微动频谱的影响。引入压缩感知方法进行干扰条件下的微动特征稀疏恢复,采用堆栈自编码学习(SAE)方法构建深层神经网络对目标进行自动特征提取和分类识别;实录数据验证表明,本文特征提取和识别方法在干扰比例41%时识别正确率能达到75%。

Aiming at the classification problem of aircraft targets in complex electromagnetic environment,this paper proposes a classification method based on compressive sensing and deep learning.The modulation features of three types of aircraft(jet aircraft,propeller airplane and helicopter)and the influence of interference are firstly analyzed theoretically and simulated.Compressive sensing and stacked sparse autoencoder are then exploited to extract modulation features under interference environment.The aircraft classification experiments are conducted by the recorded data of narrow band search radar.The results show that the correct classification rate is up to 75%even when 41%of the pulses are interfered.