k均值聚类引导的遥感影像多尺度分割优化方法

Optimization Approach for Multi-scale Segmentation of Remotely Sensed Imagery under k-means Clustering Guidance

针对不同尺度地物的分割需求,提出了一种k均值聚类引导的多尺度分割优化方法。首先对原始影像进行小尺度分割和k均值聚类,然后利用k均值聚类结果引导对象合并,在合并过程中利用Otsu阈值方法自动选择k均值聚类的影响因子,最终得到适应不同尺度地物的分割结果。以FNEA多尺度分割方法为例,利用模拟数据和真实的GeoEye-1影像数据进行相关试验,目视和定量评价表明本文方法能够得到适宜不同尺度地物的高质量分割结果。

In order to adapt different scale land cover segmentation,an optimized approach under the guidance of k-means clustering for multi-scale segmentation is proposed.At fi rst,smal l scale segmentation and k-means clustering are used to process the original images;then the result of k-means clustering is used to guide objects merging procedure,in which Otsu threshold method is used to automatical ly select the impact factor of k-means clustering;final ly we obtain the segmentation results which are appl icable to different scale objects.FNEA method is taken for an example and segmentation experiments are done using a simulated image and a real remote sensing image from GeoEye-1 satel l ite,qual itative and quantitative evaluation demonstrates that the proposed method can obtain high qual ity segmentation results.