DACIA5: a Sentinel-1 and Sentinel-2 dataset for agricultural crop identification applications

Abstract

Artificial intelligence and data analysis are essential in smart agriculture for enhancing crop productivity and food security. However, progress in this field is often limited by the lack of specialized, error-free labeled datasets. This paper introduces DACIA5, a multispectral image dataset for agricultural crop identification, complemented with Sentinel-1 radar data. The dataset consists of 172 Sentinel-2 multispectral images (800 × 450 pixels) and 159 Sentinel-1 radar images, collected over Brașov, Romania, from 2020 to 2024, with precise, in-situ verified labels. Additionally, 6,454 Sentinel-2 and 5,995 Sentinel-1 rectangular patches (32 × 32 pixels) were extracted, exceeding 6 million pixels in total. The cropland parcels considered in our dataset are used for research and are owned and cultivated by the National Institute of Research and Development for Potato and Sugar Beet, ensuring error-free labeling. The labels in our dataset provide detailed information about crop types, offering insights into crop distribution, growth stages, and phenological events. Furthermore, we present a comprehensive dataset analysis and two key use cases: crop identification based on a “past vs. present” approach and early crop identification during the agricultural season. © 2025 The Author(s). Published by Taylor & Francis Group and Science Press on behalf of the International Society for Digital Earth, supported by the International Research Center of Big Data for Sustainable Development Goals.