Anhui Province Cropland Parcels:PT-PEFT SAM: A Hierarchical Foundation Model Framework for Generalizable Agricultural and Crop Parcel Delineation

Anhui Province is situated in a unique climatic transition zone in China, encompassing a wide array of agricultural systems, from the northern plains dominated by wheat and maize to the southern hilly and mountainous regions characterized by intricate rice paddies. This environmental heterogeneity makes it an ideal yet challenging region for agricultural monitoring. Accurate, high-resolution crop parcel (CP) data is the cornerstone for precision agriculture, enabling detailed crop monitoring, resource management, and informed policy-making. This dataset provides the first high-precision, large-scale crop parcel boundaries for the entirety of Anhui Province, facilitating advanced geospatial analysis in this critical agricultural area. This crop parcel boundary dataset was developed using a novel, two-stage hierarchical deep learning framework detailed in the related publication, "PT-PEFT SAM: A Hierarchical Foundation Model Framework for Generalizable Agricultural and Crop Parcel Delineation". The process began with the generation of high-fidelity 2m resolution imagery from 10m Sentinel-2 inputs using the GeoSR super-resolution model. First, a fine-tuned foundation model (PT-PEFT SAM) was used to delineate the primary agricultural parcels (APs). Subsequently, the state-of-the-art pre-trained SAM2 model was deployed in a fully automated pipeline to perform high-fidelity segmentation of the finer crop parcels (CPs) within each AP. The dataset covers the entire province of Anhui, an area exceeding 140,100 km². "parcel.rar" data contains the vector crop parcel data in shapefile format. The files are organized and named according to the administrative divisions, following a "City_County" convention. The corresponding 2m super-resolution imagery has been compressed into two archives, image_1.rar and image_2.rar. To comply with Zenodo's 50GB file size limit per project, this record contains the first part, image_1.rar. The second part, image_2.rar, is available in a separate data record here: [https://zenodo.org/uploads/17197034].