利用弹性反馈神经网络融合ASTER GDEM和SRTM1高程数据

ASTER GDEM and SRTM1 Elevation Data Integration with RProp Neural Network

    摘要
  • 摘要: 由于数据获取与后期处理方式不同，先进星载热发射和反射辐射仪全球数字高程模型(advanced spaceborne thermal emission and reflection radiometer global digital elevation model，ASTER GDEM)和航天飞机雷达地形测绘任务(shuttle radar topography mission，SRTM)数据在高程精度上存在差异，采用弹性反馈（resilient backpropagation，RProp）神经网络算法对二者进行融合处理，实现优势互补以提升高程精度。选取两个黄土丘陵沟壑地貌样区分别用于模型建立与效果验证，1∶10 000高程精度为参考数据，在建模样区应用RProp神经网络算法构建ASTER GDEM高程校正模型、SRTM1高程校正模型、ASTER GDEM与SRTM1高程融合模型，同时应用误差反向传播(back propagation，BP)神经网络建立ASTER GDEM与SRTM1高程融合模型，将这些模型的高程精度优化效果进行对比，并在验证样区检验RProp融合模型的可行性。结果表明，RProp融合模型的高程校正效果整体上优于ASTER GDEM高程校正模型、SRTM1高程校正模型和BP神经网络融合模型，高程均方根误差分别降低6.81 m、0.34 m与0.19 m，具有良好的适用性与误差校正效果。

     

    Abstract:
      Objectives  There exist some differences between the elevation accuracy of advanced spaceborne thermal emission and reflection radiometer global digital elevation model (ASTER GDEM) and shuttle radar topography mission (SRTM) data, because their methods of data acquisition and post-processing are different.
      Methods  To improve the DEM accuracy by taking advantages of ASTER GDEM and SRTM, the resilient backpropagation (RProp) neural network algorithm is adopted to integrate them. First, we use two units in loess hilly gully topography as test areas to construct models and validate their effects. Second, we take 1∶10 000 DEM as the reference data to construct the corrected ASTER GDEM elevation model, the corrected SRTM1 elevation model and the integrated elevation model of ASTER GDEM and SRTM1 by RProp neural network algorithm. Meanwhile, the integrated elevation model of ASTER GDEM and SRTM1 is created with back propagation (BP) neural network. Then, the optimized effects of elevation precision of these models are analyzed. Finally, the model integrated by RProp is tested in the validation site.
      Results  The results show that the model integrated by RProp is better than the corrected ASTER GDEM elevation model, the corrected SRTM1 elevation model, and the model integrated by BP neural network, and its elevation deviation is reduced by 6.81 m, 0.34 m, 0.19 m, respectively.
      Conclusions  It confirms that the model integrated by RProp has good applicability and error correction effect.

     

    • HTML全文
    • 参考文献(26)
    • 相关文章
    • 施引文献
  • 资源附件(0)

/

返回文章
返回