LEVIR-CD+（变化检测）

1. 简介LEVIR-CD+ 是原始 LEVIR-CD 数据集的扩展版本，专为城市地区遥感影像变化检测任务设计，是评估变化检测（尤其是深度学习类）算法的新基准；包含 985 对超高分辨率（VHR，0.5m / 像素）的 Google 地球图像补丁，尺寸为 1024×1024 像素，相比原始版本的 637 张图像显著扩容；双颞图时间跨度达 5—14 年，聚焦建筑相关变化（含建筑增长与衰落），涵盖别墅住宅、高层公寓等多种建筑类型，由遥感图像解释专家采用二进制标签（1 表示变化，0 表示不变）标注，每个样本经双人校验确保高质量，完全注释的数据集共包含 31,333 个单独的建筑变化实例

1. Introduction: LEVIR-CD+ is an extended version of the original LEVIR-CD dataset, specifically designed for remote sensing image change detection tasks in urban areas, and serves as a new benchmark for evaluating change detection algorithms, especially deep learning-based ones. It contains 985 pairs of very high-resolution (VHR, 0.5 m/pixel) Google Earth image patches with a size of 1024×1024 pixels, which has a significantly expanded scale compared to the original dataset’s 637 images. The temporal span of the bi-temporal images ranges from 5 to 14 years, focusing on building-related changes including building growth and decline, covering various building types such as residential villas and high-rise apartments. The dataset is annotated by remote sensing image interpretation experts using binary labels (1 denotes change, 0 denotes no change), and each sample is double-checked to ensure high quality. The fully annotated dataset contains a total of 31,333 individual building change instances.

搜集汇总

背景与挑战

背景概述

LEVIR-CD+是一个用于城市地区遥感影像变化检测任务的数据集，包含985对超高分辨率Google地球图像补丁，尺寸为1024×1024像素，时间跨度为5—14年，聚焦建筑相关变化，并由专家标注。适用于基于深度学习的变化检测算法开发与评估。