Mapping global onshore wind turbines using multi-source remote sensing images and hybrid learning approaches

We developed an integrated framework that combines OpenStreetMap (OSM) data with multi-source remote sensing data (Google Earth and Sentinel-1/2) and traditional machine learning models (ResNet-18 and Random Forest) to map global onshore wind turbines. This dataset facilitates essential studies on renewable energy land management, ecological impact analysis, and data-driven energy transition policies. The dataset includes: 1. A comprehensive global inventory of 379,595 onshore wind turbines in the format of a geospatial shapefile. The dataset includes geolocation coordinates for all wind turbines, along with corresponding nation (Field: 'Nation') and land use classification (Field: 'landtype') for each wind turbine. 2. The dataset comprises 37,285 carefully annotated 256×256 pixel Google Earth image patches, containing both positive (wind turbine) and negative (background) samples, and is organized into folders with training (60%, 22,372 images) and testing sets (20%, 7,457 images) validating sets (20%, 7,456 images). The images could serve as foundational data for training deep learning models in wind turbine classification, segmentation, and detection tasks. The code file includes: 1. A PyTorch-based ResNet-18 implementation for classifying onshore wind turbines in Google Earth images, including codes for model architecture and pre-trained weights. 2. The GEE-based code for the Random Forest model, including sample point splitting (training/test sets) and model training.