基于Brushlet和RBF网络的SAR图像分类

SAR image classification using complex feature of Brushlet and RBF neural network

针对SAR图像纹理特征丰富的特点,本文提出一种新的SAR图像分类方法:通过提取Brushlet变换的能量及相位信息作为SAR图像的纹理特征,然后输入径向基函数RBF网络对图像进行分类。Brushlet变换为复值函数,具有方向信息,因此对分析富含方向信息的纹理图像十分有效,而同时提取其能量及相位特征则更优。RBF网络学习速度快,不易陷入局部极小,是一种有效的分类器。实验表明,基于Brushlet复特征和RBF网络的方法能够获得较高的分类率,性能优于传统方法。

As SAR image has rich texture feature, a new method of SAR image classification is proposed, which uses RBF neural network and the energy and phase information of Brushlet as texture feature. Brushlet transform is a complex value function with orientation information. It＇s very efficient for texture image analysis, and better to extract the complex feature. RBF network is an effective classifier for fast learning and avoiding local minimum. Using the Brushlet complex feature of energy and phase as the input of RBF network, experiments on SAR image classification show that this method outperforms the traditional methods.