msuav500k: Foundational Dataset for Multispectral and RGB UAV Imagery

The msuav500k dataset is a comprehensive collection of 598,300 curated UAV images designed for computer vision and remote sensing applications. This foundational dataset addresses critical interoperability challenges in UAV-based multispectral imaging by providing standardized, radiometrically calibrated imagery from multiple sensor platforms. Dataset Composition: 443,017 RGB images from 14 agricultural and environmental monitoring projects (2020-2025) (RGB.zip) 136,652 aligned multispectral-RGB image pairs from 7 projects (2022-2024) (RGBMS.zip) 18,629 multispectral-only images from 13 projects (2017-2022) (MS.zip) Total size: 185GB processed data Sensor Coverage: DJI platforms: Mavic 3M, Phantom 4 Multispectral MicaSense: RedEdge, Altum(PT) sensors Parrot Sequoia MAPIR Survey2 RGB Key Features: Radiometric calibration ensuring consistency across diverse sensor types Spatial alignment of multispectral bands using ORB/FLANN/RANSAC algorithms Standardized format: 512×512 pixel patches optimized for deep learning Comprehensive metadata including processing scripts and calibration details Multi-spectral bands: Blue (when available), Green, Red, RedEdge, Near-Infrared Applications: Precision agriculture (disease detection, crop monitoring, yield assessment) Environmental monitoring (forest analysis, habitat mapping) Infrastructure assessment (pavement distress, mining operations) Foundation model training and fine-tuning for remote sensing Geographic Coverage: Datasets span multiple continents with primary focus on European/Mediterranean regions (40%), including agricultural systems, vineyards, orchards, forests, and urban environments. This dataset provides researchers with a robust foundation for developing and validating computer vision algorithms, training foundation models, and conducting reproducible cross-study comparisons in UAV-based remote sensing applications. Keywords: Multispectral Imagery, UAV, Remote Sensing, Computer Vision, Radiometric Calibration, Deep Learning, Precision Agriculture