一种基于ResUNet的晕渲图生成方法

A Hill-Shading Generation Method Based on ResUNet

  • 摘要: 地貌表示的技术与方法是地图制图学的重要研究内容,地貌晕渲图是在二维平面有效表现三维地貌的主要形式。针对当前计算机晕渲图表现效果有待提升的问题,提出了基于ResUNet深度学习模型,使用数字高程模型(digital elevation model,DEM)制作晕渲图的方法,即通过构建具有对应关系的DEM和高质量手工晕渲样本集,并训练搭建的特定网络模型,以实现任意区域基于DEM数据的晕渲图制作。实验结果表明,该方法能够生成艺术性强、地貌表现效果好的晕渲图,晕渲图整体效果较大程度地接近手工晕渲作品;同时,搭建的深度学习模型适用于大多数地貌类型;与U‑Net模型相比,所提模型对不同分辨率DEM的适应性更强,且生成的晕渲视觉效果更优。

     

    Abstract:
    Objectives The representation of landform features is an important part of cartography and the relief shading is the main tool for effectively expressing the 3D landform on the two-dimensional plane. The hill-shading produced by the existing method is not artistic enough, and the expressiveness of the terrain needs to be improved. In this regard, a hill-shading generation method based on ResUNet is proposed.
    Methods First, the high-quality manual relief shading and the corresponding digital elevation model (DEM) data are processed, such as resampling, unifying coordinate system, cropping, etc., to construct the shading‐DEM sample image pairs. Then the parameters of the proposed DEM-shading conversion model are learned from the sample data. Finally, the DEM of any area is input into the model, and the output results of which are spliced using alpha blending to form the relief shading.
    Results The experimental results show that the relief shading generated by the proposed method is improved in artistry and terrain expressiveness and can closely match the quality of the manual relief shading, due to the “Swiss style”, the effective terrain generalization and the contrast optimization on both sides of the ridges. Compared with U‑Net, the proposed model is more adaptable to DEM of different resolutions, and can generated shading with better visual effects.
    Conclusions The proposed method can effectively improve the artistry and terrain expressiveness of the hill-shading.

     

/

返回文章
返回