基于特征级融合和支持向量机的飞机识别

Aircraft recognition scheme based on feature fusion and support vector machine

提出一种新的基于组合不变量的飞机识别方法。对不同飞机机型图像,提取Hu矩、仿射矩和归一化傅里叶描述子(NFD)3类不变量进行特征级融合。针对组合不变量取值范围较大问题,提出采用4种归一化方法,结合支持向量机(SVM)以提高飞机识别系统的分类性能。仿真实验表明,提取飞机的组合不变量特征,采用传统神经网络或SVM构建分类器,分类性能均优于单一类别不变量的同类分类器,且SVM的分类性能要优于传统神经网络。同时,当组合不变量要与智能型分类器结合时,采用特定的归一化方法才能取得较好的识别率。

Abstract：A new combination invariants method is proposed for aircraft recognition. For all kinds of air craft types, Hu moments, affine moments and normalized Fourier descriptors are extracted and com- bined. As the above invariants are too dispersed,four kinds of normalized methods are studied and com- bined with support vector machine （SVM） to improve aircraft classification performance. Simulation re sults show that the classification performance is better by the combination invariants which are combined with support vector machine classifier or neuron network than that by any single kind of invariants which are combined with corresponding classifiers, and the classification performance is better by support vector machine classifier than that by traditional neuron network classifier. Moreover, when combination invari- ants are sent to intelligent classifiers, a special normalization method can improve the classification performance.