The CAS Landslide Dataset is a large-scale, multisensor dataset designed for deep learning-based landslide detection. Developed by the Artificial Intelligence Group at the Institute of Mountain Hazards and Environment, Chinese Academy of Sciences (CAS), the dataset includes 20,865 RGB images from nine regions, combining satellite and unmanned aerial vehicle (UAV) data. It aims to address challenges in landslide recognition by providing a comprehensive and reliable dataset for training and benchmarking landslide identification models. The dataset includes rigorous quality control methods to ensure data integrity and effectiveness. The dataset addresses limitations of existing landslide datasets, such as small size, low spatial resolution, and insufficient diversity in landslide triggers. It includes images from various regions, including Tiburon Peninsula, Moxitaidi, Wenchuan, Palu, Lombok, Hokkaido Iburi-Tobu, Mengdong, Longxi River, Jiuzhai Valley, and Luding. The dataset includes labels created using QGIS and LabelMe, with strict standards for accuracy and quality. The dataset was built using a rigorous screening process to exclude problematic images, such as those with excessive boundary filling, low target object proportions, cloud cover, and image stitching discontinuities. The dataset is available for open access through Zenodo and includes detailed information on data sources, resolution, and subdatasets. The dataset was validated using multiple deep learning models, including FCN, U-net, DeepLabV3+, and MFFENet, showing high performance in landslide detection. It outperformed other published datasets in terms of IoU, F1 score, and mIoU. The dataset is a valuable resource for researchers to develop more accurate and efficient landslide detection models, contributing to disaster management and risk mitigation. The dataset is also useful for training and benchmarking models using multisensor data.The CAS Landslide Dataset is a large-scale, multisensor dataset designed for deep learning-based landslide detection. Developed by the Artificial Intelligence Group at the Institute of Mountain Hazards and Environment, Chinese Academy of Sciences (CAS), the dataset includes 20,865 RGB images from nine regions, combining satellite and unmanned aerial vehicle (UAV) data. It aims to address challenges in landslide recognition by providing a comprehensive and reliable dataset for training and benchmarking landslide identification models. The dataset includes rigorous quality control methods to ensure data integrity and effectiveness. The dataset addresses limitations of existing landslide datasets, such as small size, low spatial resolution, and insufficient diversity in landslide triggers. It includes images from various regions, including Tiburon Peninsula, Moxitaidi, Wenchuan, Palu, Lombok, Hokkaido Iburi-Tobu, Mengdong, Longxi River, Jiuzhai Valley, and Luding. The dataset includes labels created using QGIS and LabelMe, with strict standards for accuracy and quality. The dataset was built using a rigorous screening process to exclude problematic images, such as those with excessive boundary filling, low target object proportions, cloud cover, and image stitching discontinuities. The dataset is available for open access through Zenodo and includes detailed information on data sources, resolution, and subdatasets. The dataset was validated using multiple deep learning models, including FCN, U-net, DeepLabV3+, and MFFENet, showing high performance in landslide detection. It outperformed other published datasets in terms of IoU, F1 score, and mIoU. The dataset is a valuable resource for researchers to develop more accurate and efficient landslide detection models, contributing to disaster management and risk mitigation. The dataset is also useful for training and benchmarking models using multisensor data.