We present improvements to the physical optics method for high-frequency computations. This involves refining the calculation model such that the number of triangular facets is adequate to accurately represent the target structure. An illumination assessment is performed using the mesh model, and this is followed by a predetermined subdivision of the illuminated facets to achieve the required computational resolution. Subsequently, using a group of high-speed vehicle RAM C-II models with surfaces of different roughness, we calculate their one-dimensional high-resolution range profile and two-dimensional inverse synthetic aperture radar images at different reentry heights. The results demonstrate that the smooth-surfaced RAM C-II exhibits imaging limited to its head and tail, whereas the rough-surfaced RAM C-II reveals its entire contour. Furthermore, the presence of plasma diminishes the imaging efficacy of rough-surfaced targets, with the effect becoming more pronounced at lower altitudes. This study provides valuable insights for the identification of high-speed vehicles.