A large-scale image–text dataset benchmark for farmland segmentation

Abstract. Understanding and mastering the spatiotemporal characteristics of farmland are essential for accurate farmland segmentation. The traditional deep learning paradigm that solely relies on labeled data has limitations in representing the spatial relationships between farmland elements and the surrounding environment. It struggles to effectively model the dynamic temporal evolution and spatial heterogeneity of farmland. Language, as a structured knowledge carrier, can explicitly express the spatiotemporal characteristics of farmland, such as its shape, distribution, and surrounding environmental information. Therefore, a language-driven learning paradigm can effectively alleviate the challenges posed by the spatiotemporal heterogeneity of farmland. However, in the field of remote sensing imagery of farmland, there is currently no comprehensive benchmark dataset to support this research direction. To fill this gap, we introduced language-based descriptions of farmland and developed the FarmSeg-VL dataset – the first fine-grained image–text dataset designed for spatiotemporal farmland segmentation. Firstly, this article proposed a semi-automatic annotation method that can accurately assign captions to each image, ensuring a high data quality and semantic richness while improving the efficiency of dataset construction. Secondly, FarmSeg-VL exhibits significant spatiotemporal characteristics. In terms of the temporal dimension, it covers all four seasons. In terms of the spatial dimension, it covers eight typical agricultural regions across China, with a total area of approximately 4300 km2. In addition, in terms of captions, FarmSeg-VL covers rich spatiotemporal characteristics of farmland, including its inherent properties, its phenological characteristics, its spatial distribution, its topographic and geomorphic features, and the distribution of surrounding environments. Finally, we perform a performance analysis of the vision language model and a deep learning model that relies only on labels trained on FarmSeg-VL. Models trained on the vision language model outperform deep learning models that rely only on labels by 10 %–20 %, demonstrating its potential as a standard benchmark for farmland segmentation. The FarmSeg-VL dataset will be publicly released at https://doi.org/10.5281/zenodo.15860191 (Tao et al., 2025).