This paper reviews recent progress in remote sensing image scene classification, proposes a large-scale benchmark dataset, and evaluates state-of-the-art methods using the proposed dataset. The paper highlights the limitations of existing datasets, including small scene class sizes, limited image variations, and high classification accuracy, which hinder the development of new algorithms. The proposed NWPU-RESISC45 dataset contains 31,500 images across 45 scene classes, with 700 images per class. It offers large-scale data, significant variations in translation, spatial resolution, viewpoint, object pose, illumination, background, and occlusion, and high within-class diversity and between-class similarity. The dataset enables the community to develop and evaluate data-driven algorithms. The paper also evaluates several representative methods on the proposed dataset, including handcrafted features, unsupervised feature learning, and deep learning-based methods. The results provide a useful baseline for future research. The paper reviews existing datasets, surveys three categories of approaches for scene classification, and presents the proposed dataset in detail. It also benchmarks state-of-the-art methods on the dataset, showing their performance on a large-scale, diverse dataset. The paper concludes that the proposed dataset is a valuable resource for advancing remote sensing image scene classification.This paper reviews recent progress in remote sensing image scene classification, proposes a large-scale benchmark dataset, and evaluates state-of-the-art methods using the proposed dataset. The paper highlights the limitations of existing datasets, including small scene class sizes, limited image variations, and high classification accuracy, which hinder the development of new algorithms. The proposed NWPU-RESISC45 dataset contains 31,500 images across 45 scene classes, with 700 images per class. It offers large-scale data, significant variations in translation, spatial resolution, viewpoint, object pose, illumination, background, and occlusion, and high within-class diversity and between-class similarity. The dataset enables the community to develop and evaluate data-driven algorithms. The paper also evaluates several representative methods on the proposed dataset, including handcrafted features, unsupervised feature learning, and deep learning-based methods. The results provide a useful baseline for future research. The paper reviews existing datasets, surveys three categories of approaches for scene classification, and presents the proposed dataset in detail. It also benchmarks state-of-the-art methods on the dataset, showing their performance on a large-scale, diverse dataset. The paper concludes that the proposed dataset is a valuable resource for advancing remote sensing image scene classification.