针对飞机目标的分类问题 ,介绍了一种称为 Gabor原子网络的高分辨雷达目标距离像识别算法。 Gabor原子网络的输入层采用 Gabor原子变换作为预处理单元 ,完成对特征的提取。Gabor原子网络的隐层和输出层组成一个多层前馈网络 ,采用改进的反向传播算法对权值进行调整。文中同时给出了网络在训练过程中自动调整 Gabor原子节点的特征参数的算法。对 3种缩比模型飞机的微波暗室转台数据进行了分类 ,结果表明三维空间内的 Gabor原子网络方法比一维空间内的原始距离像或 Fourier幅度方法和二维空间内的 Gabor变换或小波变换方法更适合高分辨雷达目标距离像的识别 Aiming at the classification of aircraft targets, a high-resolution radar target distance image recognition algorithm called Gabor atomic network is introduced. The input layer of Gabor atomic network adopts Gabor atomic transform as preprocessing unit to complete the feature extraction. The hidden layer and output layer of Gabor atomic network form a multi-layer feedforward network, and the weights are adjusted by the improved back propagation algorithm. In the meantime, an algorithm for automatically adjusting the characteristic parameters of Gabor atomic nodes during training is given. The data of microwave darkroom turntables of three kinds of scale model aircraft are classified. The results show that Gabor atomic network method in three-dimensional space is better than original distance image or Fourier amplitude method in one-dimensional space and Gabor transform or wavelet transform in two-dimensional space The method is more suitable for high-resolution radar target distance image recognition