中文摘要：

与传统的基于像元的影像分类方法相比,面向对象的分类方法能够提供更为准确的地类识别结果。对象作为信息提取过程的中间实体,对其划分的好坏直接关系到影像的分类精度。为了更准确地对农区多光谱影像进行分类,提出了一种基于高精度历史耕地地块数据的影像分割方法。该方法首先判定现势遥感图像上耕地地块的均质性,然后通过计算区域对比度指标指导局部最优参数的动态选择,最终获得影像的分割结果。基于农区的案例研究表明：①在相同的全局分割参数条件下,基于高精度历史耕地地块数据的影像分割方法能够在保持稳定的耕地地块边界的同时,获得比直接影像分割更均一的对象;②局部参数的应用使得影像＂欠分割＂和＂过分割＂现象基本消除,分割结果较使用全局参数的分割结果更为合理;③局部最优分割参数的自动选取,极大地增强了该方法的客观性;④区域对比度指标对＂欠分割＂和＂过分割＂现象十分敏感,能够较好地指导影像分割,为评价农区影像分割的优劣提供了有效的评价指标。

英文摘要：

Compared with the traditional pixel-based classification method,the object-oriented classification method can reach a higher accuracy.As the middle entity in the process of information extraction,the object is one of the key factors of the object-oriented classification.The quality of the segmentation is directly related to the image classification accuracy.In this paper,a segmentation approach based on the high precision historical cropland parcels is presented.In this approach,the cropland parcels are considered to be homogeneous or not based on the real-time remote sensing image,then the global contrast index is calculated by each parcel to find the best local segmentation parameters,and the improved result is finally approached.The approach was tested in an agricultural area,and the result shows that： 1） the approach can get a more homogeneous object with stable boundaries;2） Local segmentation parameters provide a more reasonable result in which the ＂less-segmentation＂ phenomenon and the ＂over-segmentation＂ phenomenon are effectively eliminated;3） Automatic selection of local optimal segmentation parameters greatly enhances the objectivity of this approach;4） Global contrast index is sensitive to the ＂less-segmentation＂ phenomenon and the ＂over-segmentation＂ phenomenon,so it can lead the segmentation to produce the best result.On the other hand,it can also serve as a good index to evaluate image segmentation in agricultural areas.