ISPRS 城市分割遥感数据集

摄影测量的主要课题之一是从机载传感器获取的数据中自动提取城市物体。这项任务之所以具有挑战性，是因为建筑物、街道、树木和汽车等物体在高分辨率数据中的外观非常不均匀，这导致类内方差很大，而类间方差很低。重点是详细的 2D 语义分割，为多个对象类别分配标签。进一步的研究驱动因素是来自新传感器的高分辨率数据和依赖于日益成熟的机器学习技术的先进处理技术。尽管付出了巨大的努力，但这些任务还不能被视为已解决。据我们所知，目前还没有完全自动化的 2D 物体识别方法在实践中应用，尽管至少有二十年的研究试图解决这一任务。阻碍科学进步的一个主要问题是缺乏用于评估物体提取的标准数据集，因此很难通过实验比较不同方法的结果。该数据集旨在解决这个问题。

One of the core topics in photogrammetry is the automatic extraction of urban objects from data acquired by airborne sensors. This task is highly challenging, as objects such as buildings, streets, trees, and cars exhibit highly heterogeneous appearances in high-resolution data, resulting in large intra-class variance and low inter-class variance. The focus lies on detailed 2D semantic segmentation, which assigns labels to multiple object categories. Further research drivers include high-resolution data from new sensors and advanced processing technologies that rely on increasingly mature machine learning techniques. Despite considerable efforts, these tasks cannot yet be considered solved. To the best of our knowledge, no fully automated 2D object recognition method has been applied in practice to date, even though at least two decades of research have attempted to address this task. A major issue hindering scientific progress is the lack of standard datasets for evaluating object extraction, making it difficult to experimentally compare the results of different methods. This dataset is designed to solve this problem.

搜集汇总

背景与挑战

背景概述

ISPRS城市分割遥感数据集是一个用于2D语义分割的高分辨率遥感数据集，包含多个城市地区的正射影像，覆盖6种土地覆盖类别。该数据集旨在解决城市物体自动提取的挑战，适用于深度学习、计算机视觉等领域的研究。