农田遥感识别方法与应用

农田识别是进行作物的长势监测、产量预报和时空特性研究的基础,能够为国家农业政策的制定提供数据支撑。基于像元的传统农田遥感识别方法只利用了影像像元的属性信息,造成农田识别精度不高。面向对象的识别方法根据多尺度分割后得到对象的光谱、形状和纹理属性分类,提高了农田识别的精度。随着航天技术快速发展,与中低分辨率影像相比,高分辨率遥感影像能够提供更丰富的光谱、形状和纹理特征的目标地物,但同时信息量和数据量呈几何级数增加的高分辨率遥感影像给现有的农田遥感识别方法带来极大挑战。回顾了中低分辨率和高分辨率的农田遥感识别方法的研究进展,重点阐述了多尺度分割与三种监督型机器学习算法组合的面向对象的识别方法。结果表明,以机器学习算法为基础的面向对象分类的总体精度都高于96%,卡帕系数均超过0.93。

Remote Sensing Identification Method and Application of Farmland

Farmland identification is the basis of crop growth monitoring,yield prediction and spatial and temporal characteristics study,which can provide data support for the formulation of national agricultural policy.The traditional remote sensing identification method based on pixel only makes use of the attribute information of the image pixels,which causes the farmland recognition accuracy is not high.According to the classification of the spectrum,shape and texture of the object after multiresolution segmentation,the object recognition method improves the precision of the farmland recognition.With the rapid development of space technology,compared to middle /low resolution image,the high resolution remote sensing image can provide richer spectrum,shape and texture features targets,and the amount of information and data in high resolution remote sensing image presenting geometric progression increase in the farmland brings great challenges to the existing recognition methods of remote sensing.This paper reviews the research progress of middle/low resolution and high resolution remote sensing identification methods,and focuses on the combination of multi-scale segmentation and three supervised machine learning algorithms.The results show that the accuracy of object-oriented classification based on machine learning algorithm is higher than 96%,and the kappa coefficient is more than 0.93.