WorldView-3短波红外影像分类性能研究

Research on Classification Performance of WorldView-3 Short-wave Infrared Band Images

WorldView-3卫星在8个可见光-近红外(VNIR)波段的基础上,新增了8个短波红外(SWIR)波段,大大提高了对地物信息的提取能力。利用随机森林分类方法分别对可见光-近红外8个波段影像和可见光-近红外-短波红外16个波段的影像进行实验验证;采用基于多尺度分割技术的面向对象方法对合理的特征空间进行实验与挑选。结果表明,引入SWIR波段后分类性能总体精度提升了3.78%,人工地物制图精度提升了5.65%,自然地物制图精度提升了2.88%;且允许识别特定类别(村镇中的红色低矮砖瓦房居民区),能保持地物较为完整的形状信息,可提高多光谱遥感影像的分类精度。

On the basis of 8 visible-near infrared bands(VNIR),the WorldView-3 satellite adds 8 short-wave infrared bands(SWIR),which can greatly improve the ability of extracting ground information.In this paper,we used the random forest classification method to verify the eight bands images of VNIR and the SWIR bands(totally 16 bands)experimentally,and used the object-oriented method based on multi-scale segmentation technology to test and select the reasonable feature space.The results show that the overall accuracy of classification performance is improved by 3.78%,the accuracy of the artificial ground is raised by 5.65%,and the accuracy of the non-artificial ground is raised by 2.88%.Meanwhile,the specific category is allowed,such as the low red brick tile house in the rural area.It can keep shape information more complete,and be useful for improving the classication accuracy of multi-spectral remote sensing images.