A SAM-based method for large-scale crop field boundary delineation
Xuanyu Liu
Abstract
Large-scale digitalization of crop field boundaries provides vital information for smart agriculture applications. Due to the high cost of high-resolution remote sensing images and the time-consuming manual annotation of data, effective and budget-friendly solutions for extracting closed agricultural field boundaries remain scarce. Recently, the emergence of the foundation model, Segment Anything Model (SAM), has had a profound impact on the field of computer vision and prompted efforts to migrate it to particular fields. In order to explore the potential of the SAM for agricultural land segmentation, we propose a workflow that automatically instructs the SAM model for large-scale farmland division by extracting spatialtemporal features from remote sensing images as auxiliary information. The methodology was evaluated on an experimental area larger than 1,000 square kilometres, and the preliminary results corroborate its applicability and viability.