面向对象的农田信息遥感影像分割算法

An object-oriented segmentation method for RS imagery of cropland information

为了提高农业遥感数据处理中多光谱影像分割的精度,文章提出了一种面向农田信息提取的遥感影像分割算法:利用KMeans非监督分类算法和Fisher标准估算多光谱遥感影像中各个波段的权值,并将估算的波段权值应用到光谱合并计算中,能够较好地提高农田区域的分割精度,实现基于全局最优合并的区域生长算法,得到最优化的分割结果;从分割结果中提取基于区域的NDVI信息可以较为快速、准确地区分农田和非农田区域。实验结果说明:该方法的分割精度优于传统的全局最优合并算法和FNEA算法,并对遥感影像中旱田和水田的提取均有较好的效果。

In order to improve the segmentation accuracy of multi-spectral imagery in the processing of agricultural RS data, the paper proposed an object-based interpretation algorithm for agricultural RS ima- ges： the unsupervised classification algorithm KMeans, and Fisher criterion were used to estimate the weights of each band of the RS imagery, and the estimated weights were applied in the calculation of spec- tral merging, which enhanced the accuracy of cropland segmentation well; and then a global-best-mer- ging-based region growing method was implemented to obtain the optimized segmentation result, from which field-based NDVI information extracted from segmentation result produced relatively fast and accu- rately discrimination between croplands and other fields. Result indicated that the proposed method would outperform the traditional global-best-merging algorithm and FNEA, and effectively extract dry lands and paddy fields from RS images.