LCMA semantic segmentation dataset in Jiangxia

The Fine-Scale LCMA (Land Cover in Mining Area) dataset is constructed for remote sensing semantic segmentation in complex mining environments. The study area is located in Jiangxia District, Wuhan City, Hubei Province, China, covering approximately 109.4 km² (114°12′33.59′′E–114°23′6.89′′E, 30°15′38.85′′N–30°18′57.48′′N). This region has a nearly 70-year history of open-pit mining activities, with most mines still active, making it an ideal area for studying mining-induced land cover changes. The dataset is derived from a ZiYuan-3 (ZY-3) stereo satellite image acquired on June 20, 2012. The 3.6-m resolution front- and backward-looking panchromatic data were used to extract 10-m resolution relative digital terrain model (DTM) data. Subsequently, 2.1-m resolution panchromatic-multispectral fused data (16-bit) were generated and used as input features for modeling. A fine-scale land cover classification system was established by integrating the FAO classification system and the Chinese Academy of Sciences resources and environment database. The system comprises 20 classes, including seven specific to open-pit mining areas and 13 common land cover types. The ground truth was manually interpreted through field investigations and validated with 29 field survey points. For dataset construction, the study area was divided into 256×256 patches with overlapping strategies (87.1% overlap between M and L areas, 78.5% overlap between areas 31 and 32). The dataset was randomly sampled and manually fine-tuned to ensure spatial distribution balance across all areas and class distribution consistency. The final dataset contains 247 training images, 91 test images, and 78 validation images (approximately 6:2:2 ratio). All images in the dataset were normalized to zero mean and unit variance to ensure consistent input features for model training.