机载高光谱影像降维方法比较
发布时间：2018-09-10 12:59
【摘要】：高光谱数据波段多、波段之间相关性强,导致信息冗余严重,增加了数据处理的工作量,有效准确地在众多波段中选择具有代表性的波段尤为重要。首先用Wilks'Lambda(WL),随机森林(random forest,RF)与自适应波段选择(adaptive band selection,ABS)这3种方法对高光谱数据进行降维处理。然后提出了基于曲线误差指数的评价方法,用此指数的趋势来确定每种降维方法所要选择的合适波段数量,同时用指数的大小评价不同降维方法的优劣,并用分类方法对评价结果加以验证。结果显示:Wilks'Lambda最终选择的波段数为10个,α6-α平稳值(选择6个波段时的曲线误差值与曲线误差平稳值之间的差值)为0.05;随机森林最终选择的波段数为13个,α6-α平稳值为0.06;自适应波段选择方法最终选择的波段数为20个,α6-α平稳为0.14。Wilks'Lambda的总体分类精度为80.56%,Kappa系数为0.77;随机森林的总体分类精度为79.11%,Kappa系数为0.76;自适应波段选择方法的总体分类精度为49.94%,Kappa系数为0.41。得出以下结论:(1)基于曲线误差指数的方法得出Wilks'Lambda有最小的α6-α平稳值,是最佳的波段降维方法 ;分类结果显示:Wilks'Lambda有最大的总体分类精度与Kappa系数,是最佳的波段降维方法。(2)基于曲线误差指数的评价方法与基于分类结果的误差一致,说明此方法具有可行性。
[Abstract]:There are many bands of hyperspectral data and strong correlation between bands, which leads to serious redundancy of information, and increases the workload of data processing. It is particularly important to select representative bands in many bands effectively and accurately. Firstly, the hyperspectral data are reduced by using Wilks'Lambda (WL), random forest (random forest,RF) and adaptive band selection (adaptive band selection,ABS). Then an evaluation method based on curve error index is proposed. The trend of this index is used to determine the appropriate number of bands to be selected for each dimensionality reduction method. At the same time, the advantages and disadvantages of different dimensionality reduction methods are evaluated by the magnitude of exponent. The evaluation results are verified by the classification method. The results show that the final number of bands chosen by 10 bands is 10, the 伪 6- 伪 stationary value (the difference between the curve error value and the curve error stationary value) is 0.05, and the number of bands selected by random forest is 13, and the 伪 6- 伪 stationary value is 0. 05. 0.06; the number of bands selected by adaptive band selection method is 20; the overall classification accuracy of 伪 6- 伪 stationary 0.14.Wilks'Lambda is 80.56 kappa coefficient is 0.77; the total classification accuracy of random forest is 79.11kappa coefficient 0.76; the overall classification accuracy of adaptive band selection method is 0.76. The precision is 49.94 and the Kappa coefficient is 0.41. The following conclusions are obtained: (1) based on the curve error index method, Wilks'Lambda has the smallest 伪 6- 伪 stationary value, which is the best band dimension reduction method, and the classification results show that the Wilks'Lambda has the largest overall classification accuracy and Kappa coefficient. (2) the evaluation method based on curve error index is consistent with the error based on classification results, which shows that this method is feasible.
【作者单位】： 浙江农林大学浙江省森林生态系统碳循环与固碳减排重点实验室;浙江农林大学省部共建亚热带森林培育国家重点实验室;中国林业科学研究院资源信息研究所;
【基金】：国家自然科学基金青年基金资助项目(41201365) 浙江农林大学科研发展基金资助项目(2013FR052,2014FR004);浙江农林大学林学重中之重一级学科学生创新基金资助项目(201513)
【分类号】：TP751
本文编号：2234512
[Abstract]:There are many bands of hyperspectral data and strong correlation between bands, which leads to serious redundancy of information, and increases the workload of data processing. It is particularly important to select representative bands in many bands effectively and accurately. Firstly, the hyperspectral data are reduced by using Wilks'Lambda (WL), random forest (random forest,RF) and adaptive band selection (adaptive band selection,ABS). Then an evaluation method based on curve error index is proposed. The trend of this index is used to determine the appropriate number of bands to be selected for each dimensionality reduction method. At the same time, the advantages and disadvantages of different dimensionality reduction methods are evaluated by the magnitude of exponent. The evaluation results are verified by the classification method. The results show that the final number of bands chosen by 10 bands is 10, the 伪 6- 伪 stationary value (the difference between the curve error value and the curve error stationary value) is 0.05, and the number of bands selected by random forest is 13, and the 伪 6- 伪 stationary value is 0. 05. 0.06; the number of bands selected by adaptive band selection method is 20; the overall classification accuracy of 伪 6- 伪 stationary 0.14.Wilks'Lambda is 80.56 kappa coefficient is 0.77; the total classification accuracy of random forest is 79.11kappa coefficient 0.76; the overall classification accuracy of adaptive band selection method is 0.76. The precision is 49.94 and the Kappa coefficient is 0.41. The following conclusions are obtained: (1) based on the curve error index method, Wilks'Lambda has the smallest 伪 6- 伪 stationary value, which is the best band dimension reduction method, and the classification results show that the Wilks'Lambda has the largest overall classification accuracy and Kappa coefficient. (2) the evaluation method based on curve error index is consistent with the error based on classification results, which shows that this method is feasible.
【作者单位】： 浙江农林大学浙江省森林生态系统碳循环与固碳减排重点实验室;浙江农林大学省部共建亚热带森林培育国家重点实验室;中国林业科学研究院资源信息研究所;
【基金】：国家自然科学基金青年基金资助项目(41201365) 浙江农林大学科研发展基金资助项目(2013FR052,2014FR004);浙江农林大学林学重中之重一级学科学生创新基金资助项目(201513)
【分类号】：TP751
【相似文献】
相关硕士学位论文 前1条
1 唐贵华;基于密度排序聚类和超像素分割的高光谱遥感影像降维方法研究[D];深圳大学;2016年
，本文编号：2234512
本文链接：https://www.wllwen.com/guanlilunwen/gongchengguanli/2234512.html
