SemanticForestMY: a high-resolution remote sensing dataset for forest and non-forest semantic segmentation

Accurate forest and non-forest segmentation remains a challenge in environmental monitoring due to limitations in ground truth quality and dataset standardization. This study introduces SemanticForestMY, a highresolution remote sensing dataset developed from Google Earth Pro imagery across three key Malaysian ecosystems. A systematic annotation process combining automatic, assisted, and manual labeling techniques was employed to produce high-quality semantic segmentation masks. U-Net model evaluations achieved a validation accuracy of 93.10% and IoU of 91.36%, confirming dataset reliability. SemanticForestMY addresses existing dataset limitations and provides a solid foundation for future research in deep learning-based forest monitoring applications. Future work includes expanding the dataset with additional classes and temporal resolutions.