面向深度置信网络的宽带一维像预处理及识别技术

Preprocessing and Recognition of High Resolution Range Profile for Deep Belief Networks

深度学习方法具有特征自学习能力,可有效挖掘不同目标深层次、差异性特征,在目标识别领域被广为研究和应用。针对雷达一维像识别难题,本文提出面向深度置信网络的宽带一维像预处理及识别训练方法。通过对六类飞机一维像识别试验,验证了预处理、模型泛化等措施的有效性,飞机目标的全姿态平均识别正确率达88.35%,±15°的迎头姿态下识别正确率达97.44%。

Deep neural networks have the ability of self-learning.They can mine deep and distinctive characteristics among targets,and thus are widely studied and applied in the field of target recognition.To solve the problem of radar target recognition based on high resolution range profile(HRRP),a method of preprocessing and recognition for deep belief networks is proposed.Experiments on a radar HRRP dataset of six types of airplanes validate the effectiveness of preprocessing and model generalization.The overall accurate rate for the recognition of airplane target is 88.35%and the accurate rate between±15°head-on angles reaches 97.44%.