"SemanticForestMY 1.0"

"SemanticForestMY 1.0 is a spatio-temporal satellite imagery dataset designed for binary semantic segmentation of forest and non-forest land cover in Malaysia. The dataset was collected using Google Earth Pro based on Landsat-derived observations and covers three ecologically significant tropical regions: Endau Rompin (2\u00b030\u203242.20\u2033 N, 103\u00b032\u203256.39\u2033 E), Royal Belum (5\u00b047\u20329.93\u2033 N, 101\u00b030\u203251.44\u2033 E), and Tasik Chini (3\u00b022\u203232.82\u2033 N, 102\u00b036\u203211.18\u2033 E). To support long-term deforestation and land-cover change analysis, satellite images were selected from three temporal periods (1990, 2000, and 2010). All imagery is provided in PNG format with a maximum spatial resolution of 8192 \u00d7 4883 pixels, corresponding to an approximate ground coverage of 137,500 m per image. Ground truth segmentation masks were generated using an assisted annotation framework and further refined through manual correction to reduce labeling ambiguity in mixed land-cover regions. Images and masks were partitioned into non-overlapping 224 \u00d7 224 pixel tiles, producing 756 tiles per study area for each time period. The dataset is released with standardized splits consisting of 4,761 training tiles, 675 validation tiles, and 1,368 testing tiles, enabling fair benchmarking under consistent experimental conditions. SemanticForestMY 1.0 supports research in deep learning\u2013based forest mapping, land-cover monitoring, and spatio-temporal change detection using remote sensing imagery."