As one of the most well preserved polygenetic composite active volcanoes, the surface cover types in the Changbaishan Mountain volcanic eruption landform scene are widely developed and the interpretation labels in remote sensing image are clear. The remote sensing dataset is currently an important data foundation for large range volcanic eruption landform classification based on deep learning methods from remote sensing image. In this paper, a volcanic eruption landform scene dataset (VELSD 1.0) from Sentinel-2 remote sensing image for deep learning classification in Changbaishan Mountain, China is presented in combination with geological data and field investigations. In the dataset, it includes ten kinds of surfacec over types (i.e., high vegetation coverage, middle vegetation coverage, low vegetation coverage, exposed soil, basaltic rocks, trachytic rocks, mixed rocks of trachyte and pumice, water, shadow, and artificial landscape) as well as three constituent elements (i.e., remote sensing image, labeled data, interpreted label and explanatory file). Covering an area of 2500 km2, the dataset has approximately 40000 sample images of volcanic eruption landform which the image size of each sample is 25 pixels×25 pixels and the spatial resolution is 10 m. The classical convolutional neural networks (e.g., GoogLeNet, ResNet) and Transformer-based (e.g., Vision Transformer, Swin Transformer) deep learning models are used to validate and analyze the proposed VELSD 1.0 dataset. The experiments show that the proposed dataset has good applicability and transferability for deep learning models, and the overall classification accuracy (OA), Kappa coefficient (KC), and mean average precision (mAP) reached 82.93%, 75.64%, and 84.22%, respectively. It can effectively support the deep learning classification of other volcanic eruption landform scene from remote sensing image, and improve the informatization and intellectualization of remote sensing surveys for volcanic landform.