BP神经网络天线方向图获取方法研究

焦平面阵列成像体制是实现无源毫米波实时成像的主要途径。焦平面体制存在馈源偏焦的问题,导致成像系统的点扩展函数(PSF)具有空变特性。为研究毫米波焦平面阵列成像点扩展函数的空变规律,提出采用BP神经网络生成偏焦位置的空变方向图。通过分析天线方向图数据的特点,研究了BP神经网络的结构形式和输入输出数据格式的设计方法,并通过实验方法确定了网络的隐层形式。实验结果表明,采用双隐层BP神经网络生成的方向图可以达到99.75%的精度。

Antenna Pattern Generation by BP Neural Network

Focus plane array has been widely used in real-time Passive millimeter wave (PMMW) imaging. But its antenna pattern nonhomogeneity because of feed lateral displacement results in space variance of point spread function (PSF) in PMMW image.In order to study the complicated space-variant law of PSF,an approach is proposed to generate antenna patterns of offset feeds by back propagation (BP) neural network.By analyzing the characteristics of data of antenna pattern,the structure with its input/output data formats of the BP network are discussed.Experimental results demonstrate that the antenna pattern generated by the neural network with double hidden layers exhibits a precision no less than 99.75%.