SECOND (SEmantic Change detectiON Dataset)

其次是注释良好的语义变化检测数据集。为了确保数据的多样性，我们首先从多个平台和传感器收集4662对航拍图像。这些图像对分布在杭州，成都和上海等城市。每个图像具有512x512的大小，并且在像素级别被注释。SECOND的注释由地球视觉应用专家小组进行，从而保证了较高的标签精度。对于第二个数据集中的变化类别，我们关注6个主要的土地覆盖类别，即非植被地表，树木，低植被，水，建筑物和游乐场，它们经常涉及自然和人为的地理变化。值得注意的是，在新的数据集中，非植被地表 (简称n.v.g.地表) 主要对应于不透水地表和裸露土地。综上所述，这6个选定的土地覆盖类别产生了30个常见的变化类别 (包括非变化类别)。通过图像对的随机选择，第二个反映了发生变化时土地覆盖类别的真实分布。

This is followed by the well-annotated semantic change detection dataset. To ensure data diversity, we first collected 4662 pairs of aerial images from multiple platforms and sensors. These image pairs are distributed across cities including Hangzhou, Chengdu, Shanghai, and others. Each image has a resolution of 512×512 pixels and is annotated at the pixel level. Annotations for the SECOND dataset were completed by a panel of experts in geospatial vision applications, ensuring high labeling accuracy. For the change categories within this dataset, we focus on 6 major land cover classes: non-vegetated surfaces, trees, low vegetation, water, buildings, and playgrounds, which are frequently involved in both natural and anthropogenic geographic changes. Notably, in this new dataset, non-vegetated surfaces (abbreviated as n.v.g. surfaces) primarily correspond to impervious surfaces and bare land. In summary, these 6 selected land cover classes generate 30 common change categories (including the no-change category). Through random selection of image pairs, this dataset reflects the realistic distribution of land cover classes during change events.

AI搜集汇总

背景与挑战

背景概述

SECOND是一个用于遥感图像语义变化检测的数据集，包含4662对航拍图像，覆盖杭州、成都和上海等城市，图像大小为512x512并具有像素级注释。该数据集关注6个主要土地覆盖类别（如建筑物、水等），形成30个变化类别，由专家标注确保高精度，旨在模拟真实世界中的土地覆盖变化分布。