Details

Autor(en) / Beteiligte

• Martinez-Sanchez, Laura
• Hufkens, Koen
• Kearsley, Elizabeth
• Naydenov, Dimitar
• Czúcz, Bálint
• van de Velde, Marijn

Titel

Semantic segmentation dataset of Land Use/Cover Area frame Survey (LUCAS) rural landscape Street View Images

Ist Teil von

• Data in brief, 2024-06, Vol.54, p.110394, Article 110394

Ort / Verlag

Elsevier Inc

Erscheinungsjahr

2024

Link zum Volltext

Quelle

Alma/SFX Local Collection

Beschreibungen/Notizen

• Urban focused semantically segmented datasets (e.g. ADE20k or CoCo) have been crucial in boosting research and applications in urban areas by providing rich sources of delineated objects in Street View Images (SVI). However, there is a lack of similar datasets for agricultural and rural landscapes. By focusing on such underrepresented landscapes, we created a dataset containing images with visually segmented objects that were labeled following a thematically relevant set of classes. The dataset contains 1784 north-looking landscape images with their corresponding annotated masks from across Europe. Images were sourced from the Land Use and Coverage Area frame Survey (LUCAS), following a strict sampling and acquisition protocol. Objects were fully delineated on the street (eye) level or so-called landscape images for a set of 35 relevant classes (e.g. cropped fields, dense woody features, field margins, stone walls). This modest dataset, due to the cost of segmentation, might provide limitations for some applications (due to class imbalances). However, initial segmentation labels open the potential for the rapid (semi-supervised) growth of a larger dataset using LUCAS or other street level imagery. Although uncertainties remain, this annotated dataset is a first step toward integrating LUCAS image data within a landscape segmentation context. This can support land-use and land-cover change assessments, comparison and integration with Earth Observation based products, improved structural characterization of vegetation, as well as biodiversity and landscape heterogeneity monitoring. The data are structured in two folders which store the images and the masks. Also included is a csv file with the label and codes corresponding to the masks and another csv file data with geolocation information and ancillary data derived from the Harmonized LUCAS in-situ land-cover and land-use database for each image.