分组式OIF在高光谱图像分类的应用  

Application of Grouped OIF in Hyperspectral Image Classification

作　　者：周安旭 

机构地区：[1]成都理工大学数理学院,四川 成都

出　　处：《理论数学》2023年第10期2755-2763,共9页Pure Mathematics

摘　　要：高光谱图像具有高维度和高相关性,导致“维度灾难”和计算成本高。本文利用波段的标准差和相关系数,选择信息量大且相关性小的波段作为特征波段。为克服原始OIF难以在高光谱图像中采用的困境,提出分组式OIF (Grouping OIF, G-OIF)将高光谱图像分为若干子集,分别计算每个子集的最佳波段组合,然后并集得到整个图像的最佳波段组合。使用Indian Pines数据集,采用随机森林和支持向量机作为分类器,比较不同的分组和波段数对分类效果的影响。最后发现使用G-OIF时分组越多,波段数越多,分类效果越好。G-OIF能够在保证精度的同时实现降维,并缓解“维度灾难”。Hyperspectral images are highly dimensional and highly correlated, leading to the “curse of di-mensionality” and high computational cost. In this paper, the standard deviation and correlation coefficient of the bands are used to select the bands with large amount of information and low correlation as the characteristic bands. In order to overcome the dilemma that the original OIF is difficult to use in hyperspectral images, a grouping OIF (Grouping OIF, G-OIF) is proposed to divide hyperspectral images into several subsets, calculate the best band combination for each subset, and then combine Get the best band combination for the entire image. Using the Indian Pines dataset, random forest and support vector machine are used as classifiers to compare the effects of different groups and band numbers on classification performance. Finally, it is found that when using G-OIF, there are more groups, more bands and better classification effect. G-OIF can achieve dimensionality reduction while ensuring accuracy, and alleviate the “curse of dimensionality”.

关 键 词：高光谱图像 波段选择 维度灾难 图像分类 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]