一种基于多模态大模型的盲道遥感影像智能识别方法

By processing remote sensing images and standardized text data of tactile paving using a multimodal large model, the problem of insufficient utilization of semantic information in tactile paving identification was solved, enabling complete and continuous identification of tactile paving paths and improving the accuracy and reliability of identification.

CN121962963BActive Publication Date: 2026-07-17WENZHOU INST OF GEOTECHNICAL INVESTIGATION SURVEYING & MAPPING

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
WENZHOU INST OF GEOTECHNICAL INVESTIGATION SURVEYING & MAPPING
Filing Date
2026-04-01
Publication Date
2026-07-17

AI Technical Summary

Technical Problem

Existing methods for identifying tactile paving are prone to misidentification or omission in complex environments. They lack comprehensive utilization of tactile paving laying standards, spatial relationships of facilities, and road scene constraints, making it difficult to form complete and structured tactile paving identification results.

Method used

A smart identification method for tactile paving based on multimodal large model remote sensing images is adopted. By collecting and processing tactile paving remote sensing images and standardized text data, a multimodal dataset is constructed, cross-modal alignment inference and semantic reorganization are performed to generate a candidate set of tactile paving, and semantic adversarial verification is carried out to finally generate a tactile paving identification map.

Benefits of technology

It improves the accuracy of candidate area identification for tactile paving, reduces the false identification rate, and achieves complete identification and continuity verification of tactile paving paths, thereby enhancing the integrity and reliability of tactile paving path identification.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121962963B_ABST
    Figure CN121962963B_ABST
Patent Text Reader

Abstract

本发明公开了一种基于多模态大模型的盲道遥感影像智能识别方法,涉及多模态识别技术领域,包括,提取提示语输入集中的正向盲道语义特征和替代设施语义特征,分别计算盲道候选片段与正向盲道语义特征的语义匹配得分以及与替代设施语义特征的替代设施差异得分,生成语义对抗验证集;依据语义对抗验证集,推理盲道候选片段的语义一致性并进行可信度排序,生成可信盲道片段,分析可信盲道片段的语义拓扑关系并执行证据融合判定,生成盲道识别图谱。本发明通过构建多模态大模型并对可信盲道片段语义拓扑关系分析、路径组合处理以及路径拓扑推理与语义证据融合,提升盲道候选区域识别的准确性和盲道路径完整识别能力。
Need to check novelty before this filing date? Find Prior Art