講演情報
[PPS02-P12]月南極探査のための数値標高地図の超解像と標高・勾配Web API
*山田 悠太1、出村 裕英1 (1.公立大学法人会津大学)
キーワード：
超解像、月、南極探査、数値標高地図、API、EBCF-CDEM
High-resolution digital elevation models (DEMs) are important for terrain analysis and surface exploration in the lunar south polar region.
However, publicly available DEMs for this region still have limited spatial resolution.
In this study, a super-resolution framework based on the EBCF-CDEM model is applied to enhance a 5~m/pixel DEM to 2~m/pixel resolution over the entire lunar south polar region.
EBCF-CDEM represents terrain as a continuous function and enables stable large-area super-resolution using patch-based inference.
To generate a spatially continuous DEM, tile-based inference with overlap and Hann window blending is employed.
Quantitative evaluation using an independent 2~m/pixel DEM shows that elevation accuracy improves after bias correction.
A spatially varying elevation bias is identified in the large-area results and is effectively reduced using a correction based on the mode of the error distribution.
A small periodic artifact with an approximately 16-pixel period is also observed.
Analysis suggests that this artifact originates from the internal structure of the model rather than tile stitching, and its impact on terrain reconstruction is limited.
In addition, slope, aspect, and gradient vector products are generated from the original DEM, and a lightweight Web API is implemented to provide terrain information using precomputed GeoTIFF data.
These results show that the proposed framework provides a practical approach for large-area DEM super-resolution and terrain data delivery for lunar exploration.
However, publicly available DEMs for this region still have limited spatial resolution.
In this study, a super-resolution framework based on the EBCF-CDEM model is applied to enhance a 5~m/pixel DEM to 2~m/pixel resolution over the entire lunar south polar region.
EBCF-CDEM represents terrain as a continuous function and enables stable large-area super-resolution using patch-based inference.
To generate a spatially continuous DEM, tile-based inference with overlap and Hann window blending is employed.
Quantitative evaluation using an independent 2~m/pixel DEM shows that elevation accuracy improves after bias correction.
A spatially varying elevation bias is identified in the large-area results and is effectively reduced using a correction based on the mode of the error distribution.
A small periodic artifact with an approximately 16-pixel period is also observed.
Analysis suggests that this artifact originates from the internal structure of the model rather than tile stitching, and its impact on terrain reconstruction is limited.
In addition, slope, aspect, and gradient vector products are generated from the original DEM, and a lightweight Web API is implemented to provide terrain information using precomputed GeoTIFF data.
These results show that the proposed framework provides a practical approach for large-area DEM super-resolution and terrain data delivery for lunar exploration.