UAV-SCD: A High-Resolution UAV Dataset for Semantic Change Detection with Fine-Grained Annotations

Semantic Change Detection (SCD) identifies changes in land cover categories by comparing remote sensing images from various periods. The growth of the low-altitude economy has driven the use of unmanned aerial vehicles (UAVs) in acquiring remote sensing images. Compared to traditional satellite imagery, UAVs address limitations in resolution and data acquisition difficulty. UAVs provide high mobility, low cost, near-real-time data collection, and the capability to provide higher-resolution images. However, UAV imagery is also affected perspective changes, lighting differences, and complex scene interference, which increases the difficulty of change detection. Additionally, existing UAV datasets are limited, and the annotation accuracy is generally inadequate, especially the lack of fine-grained annotations for semantic change detection.To address this, we proposed UAV-SCD, a high-resolution UAV dataset for semantic change detection. This dataset covers urban and rural areas of Shaoxing, Zhejiang, and consists of 3,584 pairs of bi-temporal images, each with a resolution of 0.2 meters and a size of $512 \times 512$ pixels. It provides 7 categories of fine-grained semantic labels to facilitate more detailed change detection tasks. We evaluate the dataset using 7 advanced methods, and the experimental results show that UAV-SCD effectively supports change detection tasks, leverages the high-resolution advantages of UAV imagery, and addresses various challenges in change detection, making it of high application value.