面向对象分类方法在台风灾害信息提取中的应用研究

Application of object-oriented classification method to typhoon disaster information extraction

台风灾害是全球损失最严重的十大自然灾害之一。如何在灾情发生后,快速地进行灾情预评估与应急响应决策是目前灾害研究的重要方向。基于2013年台风"海燕"登陆前、后的菲律宾独鲁万市部分区域的卫星遥感影像,采用面向对象的方法进行了灾害检测,提取了研究区域的地物信息,并与基于像元的传统方法进行了比较。结果表明,面向对象的分类方法在台风灾害前影像分割尺度为30%,融合尺度为84%,台风灾害后影像分割尺度为15%,融合尺度为95%时分割效果最佳;最终,面向对象分类灾害前总体精度达到95.65%,Kappa系数达0.93,灾害后总体精度达到93.19%,Kappa系数达0.90;另一方面,最大似然法也优于ISODATA分类法与最小距离法,但仍不及面向对象的分类方法。

Typhoon disaster is one of the ten most serious disasters in global, how to conduct disaster pre-assess- ment and emergency response decision after the disaster is an important direction of the present disaster research. Based on two pieces of remote sensing images of the ＂Haiyan＂ typhoon about the part of Tacloban of Philippines / and using object oriented method for disaster detection, this paper extracts feature information of the research area, / and compares with traditional method based on the pixel. The results show that the best segmentation effect of the object-oriented classification method is obtained when the image segmentation scale is 30%, the fusion scale is 84% before typhoon disaster and they are 15% and 95% pespectively after typhoon disaster. Finally, object-oreint- ed classification passesses overall accuracy of 95.65 %, Kappa coefficient of 0.93 before disaster and overall accu- racy of 93.19% and Kappa coefficient of 0.90 after disaster. On the other hand,the maximum likelihood method isalso better than ISODATA classification method and minimum distance method, but it still has serious disadvantages and is not good as object-oreinted classification method.