Million-AID

在过去的几年中，遥感（RS）图像解释及其广泛应用取得了长足的进步。随着 RS 图像变得比以往任何时候都更容易获得，对这些图像的自动解释的需求越来越大。在这种情况下，基准数据集是开发和测试智能解释算法的必要前提。在回顾了遥感影像判读研究界现有的基准数据集之后，本文讨论了如何有效地准备合适的遥感影像判读基准数据集的问题。具体来说，我们首先分析了当前通过文献计量研究开发用于 RS 图像解释的智能算法的挑战。然后，我们介绍了以有效方式创建基准数据集的一般指导。根据所提供的指导，我们还提供了一个构建 RS 图像数据集的示例，即 Million-AID，这是一个新的大型基准数据集，包含用于 RS 图像场景分类的一百万个实例。最后讨论了 RS 图像标注中的几个挑战和观点，以促进基准数据集构建的研究。我们确实希望这篇论文能够为 RS 社区提供构建大规模实用图像数据集的整体视角，以供进一步研究，尤其是数据驱动的图像数据集。

In the past few years, remote sensing (RS) image interpretation and its wide-ranging applications have made remarkable progress. With RS images becoming more accessible than ever before, there is a growing demand for their automatic interpretation. In this context, benchmark datasets are a necessary prerequisite for developing and testing intelligent interpretation algorithms. After reviewing existing benchmark datasets in the remote sensing image interpretation research community, this paper discusses how to effectively prepare appropriate benchmark datasets for RS image interpretation. Specifically, we first analyze the challenges in developing intelligent algorithms for RS image interpretation through a bibliometric study. Then, we present general guidelines for creating benchmark datasets in an efficient manner. Following the provided guidelines, we also offer an example of constructing an RS image dataset: Million-AID, a new large-scale benchmark dataset containing one million instances for RS image scene classification. Finally, several challenges and perspectives in RS image annotation are discussed to promote research on benchmark dataset construction. We sincerely hope that this paper can provide the remote sensing community with a holistic perspective on constructing large-scale practical image datasets for further research, especially data-driven image datasets.

AI搜集汇总

背景与挑战

背景概述

Million-AID是一个大型遥感影像场景分类数据集，包含一百万个实例，旨在为遥感图像解释提供基准数据，以支持智能算法的开发和测试。该数据集由武汉大学、中国科学院、慕尼黑工业大学等机构于2021年联合发布，相关论文发表于IEEE期刊，适用于图像分类和场景分类任务。