论文部分内容阅读
在遥感影像分类的过程中非光谱特征起着重要的辅助作用。纹理特征作为一种重要的非光谱特征对于遥感影像分类精度的提高也有很重要的作用。以渭干河—库车河三角洲绿洲为例,利用ETM+数据,探讨了该绿洲盐渍化土地覆盖信息的提取方法。提出了基于SVM的光谱和纹理两种信息复合的分类方法,通过此方法对该绿洲进行分类研究,并将分类结果与最小距离法、最大似然法(MLC)、神经网络法(Neural net)和单源数据(光谱)SVM分类结果进行定性和定量比较分析。研究结果表明:该方法能够有效地解决单数据源分类效果破碎、分类精度不高等问题,并对高维输入向量具有较高的推广能力。总精度达到93.1795%,比单源信息的SVM分类法提高了3.1618%,比最大似然法提高了4.8252%,比神经网络法提高了7.4756%,而与最小距离法相比,总精度甚至提高了11.1029%,取得了良好的效果。与传统的分类方法的比较表明,文中所提出的分类方法具有明显的优越性和良好的前景,因此该方法更适合于遥感图像分类和盐渍化信息提取,是地物遥感信息提取的有效途径。
Non-spectral features play an important auxiliary role in the classification of remote sensing images. As an important non-spectral feature, texture features also play an important role in improving the classification accuracy of remote sensing images. Taking the Weigan-Kuqa delta oasis as an example, ETM + data was used to explore the extraction method of salinized land cover information in this oasis. This paper proposes a new classification method based on SVM for the combination of spectral and texture information. This method is used to classify the oasis. The classification results are compared with the minimum distance method, the maximum likelihood method (MLC), the neural net method, And single source data (spectrum) SVM classification results were qualitative and quantitative comparative analysis. The results show that this method can effectively solve the problem of single data source classification effect is not high, the classification accuracy is not high and so on, and has a high ability to promote high-dimensional input vector. The total accuracy is up to 93.1795%, which is 3.1618% higher than the single source information SVM classification method, 4.8252% higher than the maximum likelihood method and 7.4756% higher than the neural network method, while the total precision is even improved compared with the minimum distance method 11.1029%, and achieved good results. Compared with the traditional classification methods, the classification method proposed in this paper has obvious advantages and good prospects. Therefore, this method is more suitable for remote sensing image classification and salinization information extraction, which is an effective way to extract remote sensing information .