Line segment is the essential element of geometry objects, which contains very rich geometric information. Extracting complete and continuous line segments with semantic information from an image is of great significance for restoring the geometry structure of a scene, yet challenging. This paper proposes a multiresolution segment extraction approach, which performs semantic analysis on the line segments to distinguish the contour and the texture line segments. This approach first extracts line segments with multi-resolution thought, then combines the deep neural network technology to perform semantic analysis on line segments, and finally clusters the line segments to get the final result. In terms of line segment continuity and integrity, the proposed approach has obvious advantages compared with the commonly used line segment extraction methods. In terms of semantic analysis accuracy, the pixel accuracy of the proposed approach on the test set is achieves 97.82%.