5:15 PM - 6:45 PM
[SRD20-P03] Development of Applications for High Spatial Resolution of Satellite Spectral Data and Mineral Analysis
★Invited Papers
Keywords:Spectroscopy, Image Registration, Resource Exploration
Satellites observe the Earth's surface from an altitude of several hundred kilometers, resulting in low surface resolution, especially for optical satellites. This is particularly problematic for spectral cameras, which are often used in resource exploration, because of the trade-off between spatial resolution and wavelength resolution. Previous research has proposed approaches to increase the resolution of spectral images by fusing hyperspectral and multispectral images. However, there are some limitations in the use of the data, such as the need to acquire the same scene and the fact that the generated spectral images are composite data. To solve this problem, the authors proposed a protocol to construct high-resolution spectral data by combining spectral data with higher spatial resolution acquired at low altitudes between the satellite and the ground surface.
In this presentation, we present a new application and a use case for the analysis of high-resolution spectral images built for these high-resolution protocols. The proposed application allows processing of sporadically collected spectral images by performing only image registration, rather than merging spectral images. In addition, since the data are not merged, it is possible to use the pure spectral images that have been acquired. The low spatial resolution spectral images taken at high altitude are referred to as fixed images, and the high spatial resolution spectral images taken at low altitude (spectral images taken from lower altitudes than the fixed image at the research site) are referred to as move images. RGB images were then generated from these images, and feature point extraction was performed between the paired images. The spectral images were then image registered through the following steps: (1) feature detection, (2) feature matching, (3) transformation model estimation, and (4) image resampling transformation. Image registration was performed to match spectral images taken at higher altitudes while maintaining the spatial resolution of images taken at lower altitudes. The difference in elevation makes it difficult to automatically extract feature points because the resolution of the data differs significantly depending on the elevation difference. Because the spectral images used in this study have large differences in elevation, feature points were selected and aligned manually. In addition, a new bird's-eye view image transformation technique for spectral images was proposed to solve the positioning error caused by the shooting posture and applied to the positioning protocol.
In addition, a machine learning-based spectral processing technique was applied to the high-resolution spectral images produced by these protocols for mineral identification using a spectral analysis application developed by the authors. This study proposes an application that can be used as a basis for remote sensing technology using spectral data to identify minerals and heavy metal concentrations on the Earth's surface.
In this presentation, we present a new application and a use case for the analysis of high-resolution spectral images built for these high-resolution protocols. The proposed application allows processing of sporadically collected spectral images by performing only image registration, rather than merging spectral images. In addition, since the data are not merged, it is possible to use the pure spectral images that have been acquired. The low spatial resolution spectral images taken at high altitude are referred to as fixed images, and the high spatial resolution spectral images taken at low altitude (spectral images taken from lower altitudes than the fixed image at the research site) are referred to as move images. RGB images were then generated from these images, and feature point extraction was performed between the paired images. The spectral images were then image registered through the following steps: (1) feature detection, (2) feature matching, (3) transformation model estimation, and (4) image resampling transformation. Image registration was performed to match spectral images taken at higher altitudes while maintaining the spatial resolution of images taken at lower altitudes. The difference in elevation makes it difficult to automatically extract feature points because the resolution of the data differs significantly depending on the elevation difference. Because the spectral images used in this study have large differences in elevation, feature points were selected and aligned manually. In addition, a new bird's-eye view image transformation technique for spectral images was proposed to solve the positioning error caused by the shooting posture and applied to the positioning protocol.
In addition, a machine learning-based spectral processing technique was applied to the high-resolution spectral images produced by these protocols for mineral identification using a spectral analysis application developed by the authors. This study proposes an application that can be used as a basis for remote sensing technology using spectral data to identify minerals and heavy metal concentrations on the Earth's surface.