摘要
针对高光谱遥感测绘数据应用范围广、场景复杂、分类准确率低等问题,提出了高光谱技术的复杂场景遥感测绘数据分类方法。采用统计区域合并算法多尺度分割包含复杂场景的高光谱遥感测绘数据,降低噪声影响的同时解决阻塞问题,通过动态统计区域合并与多波段统计区域合并优化统计区域合并算法,准确获取高光谱遥感测绘数据内的主要结构。基于所获取的主要结构,采用由若干个RBM层堆叠构建深度置信网络,通过邻近像元相加求均值的方式扩充新样本,利用深度置信网络划分其类别。实验结果表明,采用所提方法分类时,数据集1的总体精度为97%,Kappa系数为0.95,而数据集2的总体精度为98%,Kappa系数为0.97,该方法可以对高光谱遥感测绘数据进行准确分类。
Aiming at the problems of hyperspectral remote sensing surveying and mapping data,such as wide application range,complex scene and low classification accuracy,a hyperspectral remote sensing surveying and mapping data classification method for complex scene is proposed.Statistical region merging algorithm is used to segment hyperspectral remote sensing surveying and mapping data with complex scenes at multi-scale to reduce the noise impact and solve the blocking problem.Through dynamic statistical region merging and multi band statistical region merging,the statistical region merging algorithm is optimized to accurately obtain the main structure of hyperspectral remote sensing surveying and mapping data.Based on the obtained main structure,a depth confidence network is constructed by stacking several RBM layers,and the new samples are expanded by adding the adjacent pixels to get the mean value,and the depth confidence network is used to classify them.The experimental results show that the overall accuracy of dataset 1 is 97%and kappa coefficient is 0.95,while the overall accuracy of dataset 2 is 98%and kappa coefficient is 0.97.This method can accurately classify hyperspectral remote sensing mapping data.
作者
崔利
秦浩然
姜海山
CUI Li;QIN Haoran;JIANG Haishan(Changchun Institute of Technology,Changchun 130021,China)
出处
《激光杂志》
CAS
北大核心
2022年第9期159-163,共5页
Laser Journal
基金
吉林省教育厅“十三五”科学技术研究规划项目(No.JJKH20191267KJ)。
关键词
高光谱技术
复杂场景
遥感测绘
数据分类
影像分割
深度置信
hyperspectral technology
complex scene
remote sensing mapping
data classification
image segmentation
depth confidence
作者简介
崔利(1981-),男,硕士,讲师,研究方向:测绘地理信息应用与开发。
