摘要
为了提高对高分辨率影像的分类精度,通过灰度差矢量法快速提取纹理特征,利用BP神经网络并辅以纹理特征,对一幅江西某地0.2m分辨率的航空影像进行分类。结果显示,对比度纹理特征能较好地反映该影像的纹理信息;对光谱特征不典型、纹理特征明显的人工树林,分类精度可达到90%以上;增加纹理特征后,影像分类的总精度也由55%提高到94%。表明这种结合纹理特征和BP神经网络的分类方法,能提高对高分辨率影像分类的精度。
For improving the classification accuracy of high resolution image, by extracting texture feature with gray level difference vector fleetly, an aerial image with 0.2m resolution from Jiangxi is classified by BP neural network with extracted texture feature.The result shows that, the contrast texture feature can describe the texture information of this image commendably; the classification accuracy over 90% for plantation with unrepresentative spectral feature and distinct texture feature; the collective classification accuracy of whole image is increased form 55% to 94% using the texture feature.It’s indicated that the classification accuracy of high resolution image can be improved using the way of combining texture feature with BP neural network.
出处
《测绘科学》
CSCD
北大核心
2008年第1期88-90,共3页
Science of Surveying and Mapping
关键词
遥感影像分类
高分辨率
纹理分析
灰度差矢量
BP神经网络
classification of remote sensing image
high resolution
texture analysis
grey level difference vector
BP neural network
