Semanticforestmy: a spatio-temporal remote sensing dataset for forest and non-forest semantic segmentation

Accurate classification of forest and non-forest regions in satellite imagery is vital for land monitoring. However, inconsistent annotations and limited standardised datasets hinder model comparability. SemanticForestMY addresses this issue with a high-resolution dataset derived from multi-temporal satellite data across three Malaysian regions. Annotations were produced using a hybrid method combining preprocessing, blob filtering, and manual correction. FCN32-VGG16 was used to benchmark performance, yielding 91.61% validation accuracy, a 94.52% F1-score, and an 89.60% IoU. These results validate the dataset's utility for deep learning segmentation. Future plans include multi-class expansion, seasonal coverage, and evaluation using advanced models.