GLH-Bridge_全球大幅面卫星影像桥梁检测数据集

GLH-Bridge是一个面向遥感影像桥梁检测的大规模数据集，包含全球多地采集的6,000张超高分辨率（VHR）遥感图像，图像尺寸从2,048×2,048到16,384×16,384像素不等，共标注59,737个桥梁目标。数据集覆盖多样地理环境和桥梁形态，每个桥梁均提供水平框（HBB）和旋转框（OBB）两种人工标注形式，弥补了现有遥感桥梁检测数据集在幅面和数据规模上的不足。该数据集支持深度学习模型在完整大尺寸影像中实现端到端桥梁检测，为跨场景算法泛化提供基准，并通过多尺度背景和长宽比差异促进鲁棒检测算法的开发。GLH-Bridge is a large-scale dataset dedicated to remote sensing image bridge detection. It comprises 6,000 ultra-high-resolution (VHR) remote sensing images acquired across multiple global regions, with resolutions ranging from 2,048×2,048 to 16,384×16,384 pixels, and a total of 59,737 manually annotated bridge instances. Covering diverse geographical environments and various bridge morphologies, each bridge in the dataset is provided with two types of manual annotations: horizontal bounding boxes (HBB) and oriented bounding boxes (OBB). This dataset addresses the limitations of existing remote sensing bridge detection datasets in terms of image scale and data volume. It enables deep learning models to perform end-to-end bridge detection on full large-size remote sensing images, serves as a benchmark for cross-scenario algorithm generalization, and promotes the development of robust detection algorithms by leveraging multi-scale backgrounds and aspect ratio variations.

搜集汇总

背景与挑战

背景概述

GLH-Bridge是一个全球性的大规模遥感影像桥梁检测数据集，包含6,000张甚高分辨率图像，标注了59,737座桥梁，覆盖多种地理环境和桥梁形态，提供水平和旋转两种标注框。该数据集旨在支持深度学习模型在大尺寸图像中进行端到端桥梁检测，弥补现有数据集在规模和多样性上的不足，并为算法跨场景泛化提供基准。