基于纹理和BP神经网络的SAR图像分类  

Study on SAR Image Classification Based on Texture and BP Neural Network

作　　者：李海权[1,2] 李春霞[3] 吴彩银[1] 胡召玲[1] 钱小龙[1] 

机构地区：[1]徐州师范大学城市与环境学院,徐州221116 [2]广西玉林市玉林高中,广西玉林537000 [3]广西师范大学教育科学院,桂林541004

出　　处：《遥感信息》2009年第3期58-63,共6页Remote Sensing Information

基　　金：江苏省高校自然科学研究计划项目(05KJB420133)

摘　　要：研究基于纹理和BP神经网络的SAR图像分类。首先用增强FROST滤波算法对SAR图像进行去噪处理。然后基于灰度共生矩阵理论提取去噪后的SAR图像多种纹理特征,并通过大量实验筛选出有效的纹理特征。最后,结合纹理特征,分别采用经典的最大似然分类法和BP神经网络分类法对SAR图像进行分类。实验结果表明:纹理信息辅助SAR图像的灰度进行分类,大大地提高了SAR图像的分类精度;基于BP神经网络的SAR图像分类精度高于最大似然分类法的分类精度。SAR image classification based on texture and BP neural network is researched in this paper. Firstly, the speckles from SAR image are eliminated by the enhanced FROST filter algorithm. Then the texture features of the denoised SAR image are extracted based on gray co-occurrence matrix, and effective texture features are screened through a large number of experi- ments. At last, combining with texture features, SAR image is classified with the classical maximum likelihood and BP neural network. The results indicate the following： considering texture information supporting, the classification accuracy of SAR im- age is greatly enhanced. The classification accuracy of SAR image based on BP neural network is higher than maximum likeli- hood method.

关 键 词：纹理 灰度共生矩阵 BP神经网络 SAR图像 最大似然法 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]