基于TSVM和纹理特征的WIS遥感影像分类方法

WIS remote sensing image classification method based on TSVM and texture features

针对宽波段成像光谱仪数据为中等空间分辨率、谱段多的特点,提出一种结合双子支持向量机(TSVM)和纹理特征的WIS遥感影像快速分类方法。选用经过几何校正的二级数据产品,进行大气校正、反射率计算等预处理后,裁切出500×500大小的影像作为实验对象。选择以相关性分析的方法获取特征波段。为解决狭小水体混合像元难以区分的问题,选择计算NDWI指数,然后提取其纹理特征。最后,基于TSVM和传统SVM进行分类实验,样本训练及预测的时间分别为14.27 s,26.41 s,得到的总体分类精度分别为87.861 3%,87.659 0%。结果表明,基于TSVM和纹理特征的分类算法不仅训练速度快,而且具有良好的泛化能力。宽波段成像光谱仪数据在中等尺度的土地覆盖分类中具有极大的应用价值。

As for the characteristics of medium spatial resolution and multi spectral coverage of broadband imaging spectrometer data,a WIS remote sensing image fast classification method in combination with the twin support vector machine(TSVM)and texture features is proposed.The secondary data products with geometric correction are selected to perform the pre⁃processing such as atmospheric correction and reflectance calculation,after which the image of the 500×500 size are cut out as the experimental objects.The feature band is obtained by means of the correlation analysis method.The NDWI index is calculated,and then the texture features are extracted to solve the indistinguishable problem of the mixed water pixels in the narrow water.The classification experimen based on TSVM and traditional SVM is carried out,the sample training and prediction time is 14.27 s,26.41 s,respectively,and the overall classification accuracy is 87.8613%,87.6590%.The results show that the classification algorithm based on TSVM and texture features not has fast training speed,but has good generalization ability.The broadband imaging spectrometer data has great application value in medium⁃scale land cover classification.