基于混合双隐层径向基函数网络的高分辨率SAR图像地物分类算法研究

本文在高分辨率条件下对传统的合成孔径雷达(SAR)图像自动地物分类技术进行了扩展研究.文章首先指出了经典的前馈神经网络模型在SAR图像地物分类中的不足,然后基于径向基神经网络(RBFN),结合混合专家系统,提出了一种变型的网络结构模型,称之为混合双隐层径向基函数网络(MDHRBFN),并将其应用于高分辨率单视单极化的SAR图像地物分类.实验结果表明,基于该模型的分类算法能够将SAR图像较好地区分为人造目标类、自然目标类、背景和阴影,具有比经典RBFN模型更好的分类效果,不但可以应用于SAR图像辅助判读,而且能够为目标识别过程提供潜在目标切片.

The High-Resolution SAR Image Terrain Classification Algorithm Based on Mixed Double Hint Layers RBFN Model

The research on the high-resolution synthetic aperture radar (SAR) image automatic terrain classification (ATC) is made.Firstly,the shortage of the traditional feed-forward neural network model in SAR image classification is concluded.Then a new model named mixed double hint layers RBFN (MDHRBFN) is presented,which combines radial basis function network (RBFN) with mixed expert system.Finally,an algorithm based on this model for the high-resolution,single-look and single-polarization SAR images terrain classification is given and evaluated.The results show that this algorithm can readily cluster the SAR image into man-made targets,natural target,background and shadow,and has better performance than the one based on RBFN model.It can not only be applied to SAR image assistant interpretation,but also offer the potential target chips for target recognition process.