一种基于原型的雷达目标识别神经网络分类器

A New Neural Network Classifier of Radar Target Identification Based on a Prototype

针对非相参雷达目标回波敏感于雷达视角和时延的特性，在基于FFT-WAVELET-MELLIN（FWM）多重变换的基础上，根据获取的特征矢量和模板的特性，运用神经网络的自适应性、鲁棒性，提出了一种基于原型的、多聚类中心的神经网络分类器NNC。经实际数据验证，在飞行目标的架次识别方面取得了满意的识别效果和实时性。文中还将NNC与BP分类器、普通多近邻分类器进行了分类性能比较。最后，对三类声纳水声目标的分类实验结果分析，证明了其广阔的应用前景。

In view of the echoes of in-coherent radar target being sensitive to its azimuth and time delay,this paper presents a neural network classifier (NNC) based on a prototype with multi-cluster, in accordance with the characteristics of vectors and templates obtained on the basis of the FFT-WAVELET-MELLIN (FWM) multi-transform and using the adaptivity and robustness of the neural network. By the available data it is proved that the described method may acquire satisfactory result with real-time in the recognition of airborne target number. Their classification and performance comparison are made among the NNC, BP classifier and normal N-neighbor classifier in this paper Also by analysing the classification experiment results of the three types of under water targets using this method, it is eventually proved that the presented method may find the prospect of a wide application.