摘要:
在覆盖玉米田的遥感图像中,玉米地块边缘区域存在大量的同物异谱现象,利用传统分割方法进行玉米地块分割时,会造成边缘出现许多非玉米小块区域,因而导致玉米种植面积统计错误。根据大面积玉米种植区域的形状分布特点,提出一种类矩形引导的玉米田种植区分割方法。首先采用最小核值相似区( smallest univale segment assimilating nucleus,SUSAN)边缘检测算子对融合后的高分一号(GF-1)卫星遥感图像进行边缘提取,然后根据闭合区域与外接类矩形的关系构建类矩形引导的相关函数,最后将类矩形阈值函数引入基于图的分割算法中实现特定形状的地块分割。将分割结果分别与基于图的分割方法、分水岭分割方法和人工解译样本进行实验比较,结果表明:本文方法能有效地分割出玉米田目标,减少了同物异谱带来的影响,分割结果更加符合玉米田实际分布特征和实际统计面积。
Abstract:
Corn field remote sensing images have a mass of end member spectral variability among marginal land area.When traditional method is used for corn block segmentation , it will produce a number of small corn plot areas at the edge and result in statistical errors of the planting area .According to the distribution characteristics of large corn planting area , an near -rectangle guided segmentation method for remote sensing images in corn field areas is proposed.First, the SUSAN ( smallest univale segment assimilating nucleus ) operator is used for edge detection from GF-1 fusion images.Then, according to the relationship between closed area and external near rectangular, the near rectangle-guided correlation function is built .At last, the near-rectangle guided threshold function is introduced into the graph -based segmentation algorithm to implement the field parcel segmentation of a specific shape .The results were compared with the graph -based segmentation algorithm , the watershed algorithm and the artificial interpretation sample . It is shown that the method proposed in this paper is effective in distinguishing different features , and the negative impact resulting from the endmember spectral variability can be reduced.The segmentation results are more in line with the actual characteristics of corn distribution , conforming with the actual statistics of the corn field area .
下载: