A Multi-Region Pléiades-Derived Dataset for Paddy Field Segmentation across Five Agro-Ecological Zones in Indonesia

Analysis-ready dataset derived from Pléiades very-high-resolution imagery for paddy field semantic segmentation across five Indonesian regions. It includes 3,108 paired 1024 × 1024 RGB image-mask PNGs (6,216 files total), with binary masks derived from official Lahan Baku Sawah (LBS) vector data, plus full-region masks, metadata, landscape metrics, and processing code. Tiles are distributed as non-georeferenced PNGs to support direct use in deep learning pipelines. The dataset supports benchmarking, few-shot learning, spatial cross-validation, and fragmentation-aware paddy mapping in tropical smallholder landscapes.