Construction of a neural network model for small-scale shaded relief maps constrained by topographic feature lines

Traditional manual methods for generating shaded relief maps can effectively highlight major topographic structures but are time-consuming and require professional skills. Analytical shading methods are faster but often lead to maps overloaded with terrain details, obscuring key topographic features, especially in Small-Scale Shaded Relief Maps (SSSR-Maps). This study focuses on the relief shading of alpine canyon terrain, introduces topographic feature lines (TFLs) as constraints, and constructs a neural network model based on Pix2pixHD, namely, TFLC-CGAN. Two generation methods, TFLC-CGAN-E and TFLC-CGAN-M, are proposed and compared. Experimental results show that TFLC-CGAN can generate SSSR-Maps with manual shading styles, simplifying terrain while preserving key features. TFLC-CGAN-E adapts better to sharply reduced TFL density, while TFLC-CGAN-M excels in feature preservation. Additionally, the relationships among digital elevation model resolution, TFL density, and the generated shaded relief map scales are explored. The proposed TFLC-CGAN offers an efficient solution for large-scale production of SSSR-Maps.