GTPBD: A Fine-Grained Global Terraced Parcel and Boundary Dataset

TL;DR

GTPBD addresses the need for a fine-grained, global terraced parcel benchmark by introducing 47,537 high-resolution images spanning 885 $km^{2}$ at 0.5–0.7 $m$ resolution, with three-level pixel, boundary, and parcel annotations. It supports semantic segmentation, edge detection, terraced parcel extraction, and unsupervised domain adaptation across three domains, enabling comprehensive cross-domain evaluation. The paper benchmarks a wide set of methods (eight SS, four ED, three APE, five UDA) using a multi-dimensional evaluation framework that combines pixel-level and object-level metrics, revealing the challenges of terraced landscapes and the limits of current models under domain shifts. This dataset provides a foundational resource for fine-grained agricultural terrain analysis and cross-scenario knowledge transfer, with potential impacts on land governance, food security, and erosion monitoring, while also outlining limitations and future expansion paths such as multimodal data integration and broader regional coverage.

Abstract

Agricultural parcels serve as basic units for conducting agricultural practices and applications, which is vital for land ownership registration, food security assessment, soil erosion monitoring, etc. However, existing agriculture parcel extraction studies only focus on mid-resolution mapping or regular plain farmlands while lacking representation of complex terraced terrains due to the demands of precision agriculture.In this paper, we introduce a more fine-grained terraced parcel dataset named GTPBD (Global Terraced Parcel and Boundary Dataset), which is the first fine-grained dataset covering major worldwide terraced regions with more than 200,000 complex terraced parcels with manual annotation. GTPBD comprises 47,537 high-resolution images with three-level labels, including pixel-level boundary labels, mask labels, and parcel labels. It covers seven major geographic zones in China and transcontinental climatic regions around the world.Compared to the existing datasets, the GTPBD dataset brings considerable challenges due to the: (1) terrain diversity; (2) complex and irregular parcel objects; and (3) multiple domain styles. Our proposed GTPBD dataset is suitable for four different tasks, including semantic segmentation, edge detection, terraced parcel extraction, and unsupervised domain adaptation (UDA) tasks.Accordingly, we benchmark the GTPBD dataset on eight semantic segmentation methods, four edge extraction methods, three parcel extraction methods, and five UDA methods, along with a multi-dimensional evaluation framework integrating pixel-level and object-level metrics. GTPBD fills a critical gap in terraced remote sensing research, providing a basic infrastructure for fine-grained agricultural terrain analysis and cross-scenario knowledge transfer.

Figures (17)

• Figure 1: Comparisons between previous dataset that focus on regular plain farmlands with limited terraced fields, and our GTPBD that offers more challenging and comprehensive evaluations.
• Figure 2: The characteristics of our proposed GTPBD with diverse terrain, fine-grained annotation and multi-task support. GTPBD covers seven major geographic zones in China and transcontinental climatic regions around the world.
• Figure 3: Statistics for the number and area of GTPBD dataset. (a) Distribution of images across different regions. (b) Distribution of images across different dataset splits. (c) Distribution of area across different regions. (d) Distribution of the parcel sizes (logarithmic scale in vertical axis).
• Figure 4: Statistics for three different domains. (a) Distribution of area across different domains. (b) The number of images across different domains. (c) Spectral statistics of mean and standard deviation ($\sigma$) for different domains. (d) Distribution of parcel sizes across different domains (logarithmic scale).
• Figure 5: Some cases for three different domains: Global, North and South.