Coal site related datasets

The constructed dataset comprises two representative scene types. The open-pit coal mining scenes (OCMS) include 201 original samples, which yielded 3247 standardized 224×224-pixel samples after target purification. The dataset is split 8:1:1 into training (2597 samples), validation (349 samples), and testing (301 samples). The composite coal-related scenes (CCS) include 272 original samples, which yielded 1531 standardized samples, also divided 8:1:1 into training (1223), validation (154), and testing (154). The difference in sample size reflects the varying complexity of the two scene types: OCMS scenes contain more extractable target regions, whereas CCS scenes are relatively scattered in spatial distribution. This dataset construction approach accounts for the unique characteristics of each scene, ensuring an adequately balanced and representative sample foundation for subsequent model training.