基于稀疏重构和CNN的转发干扰检测方法

Forward interference detection method based on sparse reconstruction and CNN

针对单个距离环样本未采样到转发干扰,仅稀疏重构了杂波部分而导致机载雷达空时自适应处理性能严重下降的问题,研究对比了两种基于卷积神经网络(CNN)的干扰检测方法。一种基于经典的LeNet-5模型,另一种采用较新的GoogLeNet模型。利用杂波和干扰在角度-多普勒维平面上的不同分布特性,通过对稀疏重构后的空时功率谱进行识别分类,测试结果显示识别准确率可达84%和99.8%。进而仅利用采样到干扰的训练样本,最终稀疏重构出正确的杂波与干扰空时功率谱。仿真结果验证了该方法的有效性。

Aiming at the problem that a single range loop sample is not sampled to the forwarding interference,and only the clutter part is sparsely reconstructed,which results in a serious degradation of the airborne radar space-time adaptive processing performance,the study compared two types based on convolutional neural networks(CNN)Of interference detection methods.One is based on the classic LeNet-5 model,and the other uses the newer GoogLeNet model.Using the different distribution characteristics of clutter and interference on the angle-Doppler dimension plane,by identifying and classifying the space-time power spectrum after sparse reconstruction,the test results show that the recognition accuracy can reach 84%and 99.8%.Furthermore,only the training samples sampled to interference are used to finally reconstruct the correct clutter and interference space-time power spectrum sparsely.The simulation results verify the effectiveness of the method.