摘要
遥感影像有着丰富的纹理信息,对于遥感影像分类精度的提高有着重要的作用。本文利用空间统计学中的变差函数模型来提取遥感影像的纹理信息,通过实例对提取的方法进行了研究,并通过不同变异方向纹理图像的分析比较,阐述了纹理特征提取过程中选取的三个因子、不同计算方向对纹理图像生成结果的影响,实验和分析表明了变差函数法是遥感图像纹理特征提取的一种有效手段。最后将成功提取的纹理信息与光谱信息相结合,运用BP神经网络进行分类和地物识别。实验部分将基于光谱信息的最大似然分类和基于光谱和纹理信息相结合的最大似然分类、基于光谱信息的神经网络分类和基于光谱和纹理信息相结合的神经网络分类四种分类结果进行精度对比分析,结果表明基于光谱和纹理信息相结合的神经网络分类精度最高,验证本论文提出的基于变差函数和神经网络的遥感影像分类方法是可行的。
There is abundant texture information in remote sensing image. The texture information participate in has the vital effect to enhance the remote sensing image classification precision. This study uses the variogram model in spatial statistics to extract textural information of the remote sensing image, studies the extraction method by example, analyses and compares texture image with different variations direction, expatiates the influence of texture image by select the three factors, different calculation direction. The experiment and analysis shows that variogram is an effective means to extract remote sensing image texture characteristics. Finally？ adopt the BP neural network method to classify and recognize the matter combining the textural information and spectral information, The experiment part contrasts the four results that maximum likelihood classification based on spectral information, maximum likelihood classifica- tion based on textural information and spectral information, the neural network classification based on spectral information, the neural network classification based on textural and spectral information, The results indicate that the precision of the neural network classification based on textural and spectral information is the best. The results illuminate that the neural network classification based on textural and spectral information is doable.
出处
《遥感信息》
CSCD
2009年第1期60-65,共6页
Remote Sensing Information
关键词
遥感影像分类
光谱信息
纹理信息
变差函数
神经网络
remote sensing image
spectral information
textural information
variogram
the neural network
