A novel method for generating pseudo-measured data tailored for inverse synthetic aperture radar (ISAR) learning-based imaging is proposed in this paper. This method initially performs semantic segmentation on images containing target categories from publicly available datasets to extract the geometric contour of the target. Subsequently, the geometric contour is meshed and mapped to a Cartesian coordinate system through a gridding process. A random terrain generation algorithm is then employed to randomly generate scattering blocks within the target contour, completing the data generation process. This approach not only simulates the basic shape of the target but also generates a random scattering distribution within the geometric contour that closely resembles real measurement data. As a result, it provides richer and more realistic image data for data-driven ISAR imaging methods. Experimental results indicate that the pseudo-measured datasets produced using this method, under the SRCNN model, yield superior imaging results in terms of peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM) compared to those obtained using publicly available datasets.