The invention provides a method of intelligent remote-sensing image understanding of describing ground-object space relationship semantics. The method includes: constructing a remote-sensing image semantics understanding reference library, and carrying out intelligent semantics description of a remote-sensing image. The step of constructing the remote-sensing image semantics understanding reference library includes acquiring contents such as four parts of remote-sensing image blocks, vector data blocks corresponding to the remote-sensing image blocks, description sentences corresponding to theremote-sensing image blocks, and targets in remote-sensing images and relationship graphs among all the targets. The step of carrying out intelligent semantics description of the remote-sensing imageincludes: training a model, testing the model, and carrying out intelligent remote-sensing image description, which is based on ground-object space relationships, on a remote-sensing image block outside a database. Effects of applying the technical solution of the invention are that: understanding on the remote-sensing image is realized from different angles; the established remote-sensing imagehigh-level semantics reference library allows later further addition of description sentences corresponding to remote-sensing images, and improvement of ability of image understanding and descriptionof a computer is facilitated; and a processing method is streamlined, and realization is easy.