Aerial Imagery and Semantic Segmentation Masks Dataset for Wildfire Risk Identification

This dataset contains high-resolution aerial and satellite imagery paired with pixel-level semantic segmentation masks. It was developed to train and evaluate deep learning models (such as U-Net architectures with ResNet backbones) for identifying and classifying forest fire risk zones based on environmental and visual patterns, including vegetation density and combustible organic matter. Dataset Structure The repository is structured into two main subfolders, ensuring an exact 1:1 alphanumeric pairing between the source scenes and their corresponding ground-truth labels: Images: Original aerial and satellite image patches in .png format (extracted via Google Satellite/Landsat 8 from Mexico). Masks: Corresponding pixel-level categorical segmentation masks in .png format. Class Dictionary The semantic segmentation masks contain discrete pixel values representing the initial level of wildfire susceptibility: Class 0 (Value 0): Low Risk Class 1 (Value 1): Medium / Moderate Risk Class 2 (Value 2): High Risk Context & Application This dataset serves as the visual backbone for hybrid risk assessment architectures. The categorical maps generated from these images are intended to be integrated into hierarchical Fuzzy Inference Systems (FIS) alongside real-time meteorological variables (temperature, relative humidity, and wind speed) to dynamically scale and adapt fire propagation risk indices. To facilitate data exploration, a Python script (visualize_dataset.py) is included in the root directory to automatically pair images with their masks and generate comparative visual overlays.