Annotated Image Dataset for Rainfall-Induced Clustered Landslides

Heavy rainfall often triggers a large number of mass landslides with high spatial density and wide impact range, posing a serious threat to human life safety and infrastructure. Quickly and accurately identifying such landslides is of great significance for emergency response and risk assessment. In recent years, deep learning technology has shown significant advantages in landslide automatic detection based on remote sensing images. However, there is still a lack of high-quality annotated datasets for mass landslide events induced by heavy rainfall, which limits the improvement of model performance and the validation of generalization ability. To compensate for this deficiency, this study constructed a dataset of rainfall induced landslides based on high-resolution remote sensing images. The dataset selected three typical heavy rainfall induced landslide events in central and eastern China in 2024 as the research objects, located in Jiangwan, Shaoguan, Guangdong, Wuping, Longyan, Fujian, and Zixing, Chenzhou, Hunan. The data source is PlanetScope satellite remote sensing images from the official Data Explorer platform of Planet, with a spatial resolution of 3 meters. Landslide interpretation follows a unified standard, using a fixed step size to label the landslide body, without including image geometric correction, ensuring the originality and consistency of the data. After pruning and filtering, the dataset is complete and has no missing records, and no positioning errors have been introduced. To ensure annotation accuracy, field investigations were conducted in some typical areas, and high-resolution aerial photographs were collected using drones to verify the interpretation results. The final dataset contains 3297 images of 512 × 512 pixels, with an average of approximately 19 landslides per image. The dataset contains three types of files: img (cropped raw remote sensing images), label (manually interpreted landslide annotations), and mask (mask files suitable for deep learning model training and validation), all in TIFF format. This dataset provides important data support for intelligent identification, disaster assessment, and model generalization research of mass landslides induced by heavy rainfall.