基于GF2号卫星影像的农业信息提取方法对比分析

Comparison and Analysis of Agricultural Information Extraction Methods Based upon GF2 Satellite Images

以GF2卫星0.8 m全色/3.2 m多光谱分辨率遥感影像为基础数据源,对基于GF2号卫星影像的农业信息提取流程和方法进行了研究与对比分析。首先对GF2号卫星影像进行波谱分析;其次对GF2号影像进行融合,并对多种融合方法进行质量评价;最后选择阈值法、波谱间关系法、非监督分类法和面向对象法分别对GF2号影像数据进行农业信息提取试验,并对信息提取结果进行精度验证和结果分析。试验表明,面向农业信息提取的GF2号卫星影像融合方法中,Pansharp融合算法融合影像色彩正常,无虚影,清晰度高,地类对比度正常,纹理清晰,熵值及与原始多光谱影像的相关系数高。阈值法和谱间关系法适用于提取单要素农业信息,非监督分类法能够初步获取研究区土地利用情况,面向对象法提取研究区全要素信息精度高。总体来说,不同信息提取方法具有各自的优势,在具体实际应用中,可以根据目标地类的波谱特性,选择适宜的遥感影像处理和信息提取方法。

The study chooses 0. 8 m panchromatic / 3. 2 m multi spectral resolution GF2 remote sensing image as the basic data.Information extraction processes and methods based upon GF2 resolution satellite images are compared and analyzed. In order to obtain spectral characteristics of different types of information,the spectrum of GF2 satellite image is analyzed. Then,the spectral and spatial information of remote sensing images are fused. Furthermore,the study compared quality evaluation of different fusion methods. Then,the study makes the agricultural information extraction experiments by using the threshold,the relationship between the spectral method,unsupervised classification and object-oriented method. The paper also verifies the accuracy of information extraction results and takes results analysis. Tests showed that Pansharp fusion algorithm was the best method among these methods,it has the advantage of normal color image,no ghost,high definition,class contrast to normal,clear texture,high entropy,high correlation coefficient of fusion images and the original multispectral images. Threshold and spectral relations method applies to extract a single element of agricultural information. Unsupervised classification method is suitable for obtaining land use in the study area quickly and preliminarily.The method of object-oriented has the high precision for whole information extracted. Overall,the different information extracted method has its own advantages. In practical,the people can select the best method according to the spectral characteristics of the class.