Spectral Reconstruction Method of Mid-Infrared Surface Characteristics Based on Non-Negative Matrix Factorization
LI Yin-na1, 2, LI Zheng-qiang1, 2*, ZHENG Yang1, HOU Wei-zhen1, 2, XU Wen-bin1, 3, MA Yan1, FAN Cheng1, GE Bang-yu1, YAO Qian1, 2, SHI Zheng1, 2
1. Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing 100101, China
2. University of Chinese Academy of Sciences, Beijing 100049, China
3. Science and Technology on Optical Radiation Laboratory, Beijing Institute of Environmental Characteristics, Beijing 100854, China
Abstract:In the field of mid-infrared remote sensing, hyperspectral surface reflectance/emissivity has high application value and application demand. However, it is difficult to obtain hyperspectral surface reflectance/emission characteristics in absorption band by satellite remote sensing, and there are still many problems in the method of obtaining full band surface characteristics by spectral reconstruction. In order to solve the problem of mid-infrared full band spectral reconstruction of land surface characteristics, based on the Johns Hopkins University (JHU) surface spectrum library and moderate resolution imaging spectrometer (MODIS) short wave infrared and mid-infrared surface multi spectral satellite products, A method for hyperspectral surface reflectance reconstruction using nonnegative matrix factorization (NMF) is proposed. The spectral resolution of the reconstructed hyperspectral reflectance/emissivity in the mid-infrared range of 2.5~5.0 μm can reach 10 nm. Firstly, four typical types of ground objects (soil, vegetation, artificial materials and rocks) were selected based on the JHU spectral library to establish the sample information of surface features. Then, using the spectral response function of the MODIS sensor, the reflectance results of 2.0~5.0 μm band were resampled to 301 bands with 10 nm spectral interval according to the equivalent calculation formula to obtain the JHU surface reflectance spectrum data set. Four endmember vector spectral curves were extracted by non-negative matrix decomposition of the spectral data set. Combined with the global monthly average surface reflectance/emissivity products of MODIS short wave infrared and mid infrared bands (2.13, 3.75, 3.96 and 4.05 μm), the weight coefficient vector corresponding to each pixel can be calculated, and the spectrum reconstruction of any band can be carried out to obtain the global land 5 km×5 km resolution of the monthly mean surface reflectance reconstruction results. At the same time, in order to comprehensively evaluate the spectral reconstruction method, the sub datasets of MODIS short wave infrared and mid infrared bands (2.13, 3.75, 3.96 and 4.05 μm) were extracted from the spectral data set, and the corresponding weight coefficient vector results were calculated, and the full band reflectance spectral reconstruction in the spectral range of 2.0~5.0 μm was performed. The average absolute error and relative error of the reconstruction results are better than 0.01 and 10%, respectively, which can meet the accuracy requirements of spectral reconstruction under the condition that only MODIS satellite data of 4 bands are available. In order to meet the needs of visualization of reconstruction results, based on WebGIS (Web Geographic information system, WebGIS) technology, using cesium framework and browser/server architecture, a two-dimensional and three-dimensional integrated visualization system was built, which integrated satellite base map, terrain data and spectral reconstruction results, to conduct intuitive multi-factor analysis, It provides support for the demonstration and verification of satellite products.
李殷娜,李正强,郑 杨,侯伟真,徐文斌,马 䶮,樊 程,葛邦宇,姚 前,史 正. 基于非负矩阵分解的中红外地表特性光谱重建方法[J]. 光谱学与光谱分析, 2024, 44(02): 563-570.
LI Yin-na, LI Zheng-qiang, ZHENG Yang, HOU Wei-zhen, XU Wen-bin, MA Yan, FAN Cheng, GE Bang-yu, YAO Qian, SHI Zheng. Spectral Reconstruction Method of Mid-Infrared Surface Characteristics Based on Non-Negative Matrix Factorization. SPECTROSCOPY AND SPECTRAL ANALYSIS, 2024, 44(02): 563-570.