Fusion semantic segmentation of urban heritage spatial visual level extraction algorithm

Abstract

This invention discloses an algorithm for extracting the visual hierarchy of urban heritage space by fusing semantic segmentation, comprising the following steps: S1: Multi-view image data acquisition and preprocessing: Acquiring a high-resolution, high-overlap multi-view image set of the target urban heritage space and preprocessing it. A unified deep learning model is used to process each image in the multi-view image set, generating corresponding semantic label maps and monocular depth maps respectively. This invention can not only identify all buildings, but also clearly quantify the absolute dominance of the Hall of Supreme Harmony as the "main layer," and how the Hall of Literary Profundity and the Hall of Righteousness on both sides, as the "secondary layers," enhance its majesty, and how the Meridian Gate and the sky in the distance, as the "background layer," define its magnificent field. This shift from "recording appearances" to "deep mechanistic analysis" is a key step in heritage research moving from "descriptive science" to "interpretive science."