Keywords
Mạng nơ ron Hopfield
Nâng cao độ phân giải DEM
Hopfield Neural Network
Downscaling
Working Languages
How to Cite
Abstract
The paper proposes a method to accurately approve the digital elevation model (DEM) using additional point elevation data. Although the resampling methods such as bilinear, bi-cubic, Kriging and especially the Hopfield Neural Network downscaling showed the improvement in accuracy of the DEM, the additional data such as point elevation data is useful for DEM's accuracy increase for the free global DEM data such as SRTM, ASTER and so on. The proposed approach used a new elevation gain function and a modified elevation constraint for the HNN. This newly proposed model was tested using SRTM 30 m DEM in Cao Bang, Vietnam, in an area of 1650 m × 1344 m with 130 elevation points for accuracy improvement and 64 elevation points for validation. The result showed that the accuracy has increased by 30% regarding Root Mean Square Error (RMSE) compared with the original DEM and the downscaled DEM by HNN (without additional elevation data). It means that the new methods can be applicable after further evaluation