高光谱图像辐射位深残差量化及其对地物分类影响分析  

Residual Quantization of Radiation Depth in Hyperspectral Image and Its Influence on Terrain Classification

作　　者：王娟 张爱武[1,2,3] 张希珍[1,2,3] 陈云生 WANG Juan;ZHANG Ai-wu;ZHANG Xi-zhen;CHEN Yun-sheng(Key Laboratory of 3D Information Acquisition and Application,Ministry of Education,Capital Normal University,Beijing 100048,China;Engineering Research Center of Spatial Information Technology,Ministry of Education,Capital Normal University,Beijing 100048,China;Geographical Environment Research and Education Center,Capital Normal University,Beijing 100048,China)

机构地区：[1]首都师范大学三维信息获取与应用教育部重点实验室,北京100048 [2]首都师范大学空间信息技术教育部工程研究中心,北京100048 [3]首都师范大学地理环境研究与教育中心,北京100048

出　　处：《光谱学与光谱分析》2024年第3期872-882,共11页Spectroscopy and Spectral Analysis

基　　金：国家自然科学基金项目(42071303,41571369);青海省科技成果转化专项项目(2022-NK-136);北京市教委—市自然基金联合资助项目(KZ202110028044)资助。

摘　　要：目前的研究大多集中在高光谱图像(HSI)的空间和光谱分辨率的提升及应用上,很少关注辐射分辨率的综合运用。辐射分辨率反映传感器接收地物辐射能量动态变化的范围,探测地物辐射能量的微小变化,同样包含着丰富的地物信息。该研究提出了一种高光谱图像辐射位深残差量化(HSI radiation bit depth residual quantization method)方法,构建出高光谱图像不同辐射位深层级的位深特征图像(LHSI)及其残差图像(RHSI),并通过实验综合运用高光谱不同辐射位深层级的位深特征图像和残差图像及其组合进行地物分类,并分析其对地物分类精度的影响。实验表明,在保证一定分类精度的基础上,辐射位深为9 bit的位深特征图像,保留了原始高光谱图像的主要信息;辐射位深为4 bit的残差图像,比原始高光谱图像更突出地物细节信息;13 bit的位深特征图像与3 bit的残差图像的组合,既能保留原始高光谱图像的主要信息又能突出地物细节。Most of the current research focuses on the improvement and application of spatial and spectral resolution of Hyperspectral Image(HSI).It pays little attention to the comprehensive application of radiation resolution.The radiation resolution reflects the range of the dynamic change of the radiation energy received by the sensor.It detects the small change of the radiation energy of the ground object,which also contains rich ground object information.This study proposes a HSI Radiation Bit Depth Residual Quantization Method to construct Low Bit Depth Hyperspectral Image(LHSI)and Residual Hyperspectral Image(RHSI)with different radiation bit depth levels.Through experiments,LHSI and RHSI of different radiation bit depth levels of HSI and their combinations are used to classify ground objects,and their effects on the classification accuracy of ground objects are analyzed.Experiments show that,based on ensuring a certain classification accuracy,9-bit LHSI retains the main information of HSI;4-bit RHSI highlights more details of ground objects than the HSI.The combination of 13-bit LHSI and 3-bit RHSI can not only retain the main information of HSI but also highlight the details of the ground object.

关 键 词：高光谱图像 辐射分辨率 辐射位深残差量化 地物分类 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]