基于高光谱图像的目标特征提取方法研究

发布时间：2018-04-08 22:29

本文选题：高光谱图像　切入点：主成分分析　出处：《长春理工大学》2017年硕士论文

【摘要】：近几十年,高光谱成像技术迅猛发展,尤其在遥感探测等领域有着越来越广阔的前景。对于高光谱图像数据的处理,本文提出了一种将图像融合算法和决策树算法两者相结合的目标特征提取方法。首先使用公用图像来探索目标特征提取方法,发现对于所选用的公用图像,使用Gram-Schmidt方法得到的图像融合效果要优于使用主成分分析算法融合的,接着使用融合效果好的图像,配准后与决策树分类方法结合使用,成功将图像中的山峦、河流、植被等光谱信息提取出来了。根据实验室现有的条件,对两组实验样本验证这种方法,对第一组实验样本发现,只使用主成分分析就可以将目标的绝大多数信息提取,效果不错。对第二组实验样本,则使用主成分分析和决策树分类结合的方法对其进行提取,成功将目标分离。总结经验发现:针对不同实验样本,将图像融合算法和决策树算法两者相结合均可以将目标分离提取,分离效果较好,该方法普遍适用于高光谱图像的目标特征提取,为进一步处理图像数据做出了一定的贡献。
[Abstract]:In recent decades, hyperspectral imaging technology has been developed rapidly, especially in the field of remote sensing detection and has more and more broad prospects.For the processing of hyperspectral image data, this paper proposes a target feature extraction method which combines image fusion algorithm and decision tree algorithm.Firstly, the common image is used to explore the target feature extraction method. It is found that for the selected common image, the image fusion effect obtained by using Gram-Schmidt method is better than that by using principal component analysis algorithm, and then the image with good fusion effect is used.After registration and decision tree classification, the spectral information of mountains, rivers and vegetation in the image is extracted successfully.According to the existing conditions in the laboratory, this method is verified for two groups of experimental samples. For the first group of experimental samples, it is found that most of the information of the target can be extracted by using principal component analysis only, and the effect is good.For the second group of experimental samples, the method of combining principal component analysis and decision tree classification is used to extract the target, and the target is separated successfully.Summing up experience, it is found that the combination of image fusion algorithm and decision tree algorithm can separate the target from each other, and the separation effect is good. This method is generally suitable for the target feature extraction of hyperspectral images.It has made some contribution to further processing image data.
【学位授予单位】：长春理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP751

【参考文献】