面向对象的遥感影像最优分割尺度监督评价

Supervised evaluation of optimal segmentation scale with object-oriented method in remote sensing image

面向对象的遥感影像分类质量和精度,不仅取决于分类算法的好坏,而且取决于遥感影像的分割质量。以定量方法确定最优分割尺度,排除主观因素干扰,已成为影像分割质量评价的重点。以往的分割质量评价方法往往忽视了对象识别在影像分割质量评价中的重要性,因此,在分析地表真实地物和影像分割对象之间空间关系的基础上,构造出一种基于面积和位置的影像分割最优尺度评价指数;并对World View2多光谱影像进行分割实验,确定了不同地物的最优分割尺度。研究结果表明,该方法在影像分割结果评价和参数优化方面具有更大的优势,不仅可以评价遥感影像分割质量、进行分割尺度参数优化,而且在分割质量评价过程中减少了人为干预,提高了方法的客观性。

The object -oriented classification classification algorithm but also on the goodness quality of the remote sensing images depends not only on the of the segmentation results. The quality of image segmentation determines the accuracy of subsequent classification of the remote sensing images. The quantitative method for determining the optimal segmentation scale and eliminating the interference of subjective factors becomes the focus of the image segmentation quality assessment. However, the importance of object recognition in image segmentation quality evaluation is often ignored in the previous segmentation quality evaluation method. After analyzing the complex spatial relations between the image objects and the actual image region, a new optimal segmentation scale evaluation index based on the area and position of the image object was proposed to evaluate the optimal segmentation scale. Based on the evaluation index, a WorldView2 muhispectral image was used to be researched and the optimal segmentation parameters were determined. The results show that the segmentation scale evaluation index is effective in image segmentation quality assessment and parameter optimization. The experimental results have also shown the effectiveness of the method proposed in this paper for both segmentation quality assessment and optimal parameter selection. Also, the procedure of segmentation quality assessment can be conducted with less human intervention, making the result more objective.