The DLR HedgeRows are a vector data set representing hedges in Bavaria. In total, more than 560,000 objects were recorded, with each hedge representing a single polygon (multi).
The analysis is based on RGB aerial images taken between 2019 and 2021 and in-situ hedgerow polygons. These are used together in a DeepLabV3 Convolutional Neural Network (CNN). The CNN has a Resnet50 backbone and was optimized with the Dice loss as a cost function. The generated hedge probability tiles were post-processed by merging and averaging the overlapping tile boundaries, simplifying the shape and filtering (only hedges with an area >70m² were retained). The hedge vector shapes were clipped to the boundaries of Bavaria.

The dataset is provided in the format of a GeoPackage (.pgkg) in the ETRS 1989 UTM (Zone 32N) projection (EPSG:25832) and covers the whole of Bavaria.

The layer has two attributes, a feature ID (fid) and the acquisition date (acquiDate). This attribute specifies the acquisition date of the RGB aerial image that was used to recognize the respective hedge object. Since the Bavaria-wide image acquisition was carried out in the period from April 2019 to September 2021, this field allows the assignment of an exact observation date of each individual hedge.  

The HedgeRows dataset is licensed under CC-BY-4.0.

When using this data set, please refer to: 
Huber Garcia, V.; Kriese, J.; Asam, S.; Kerler, K.; Buchner, J.; Dirscherl, M.; Stellmach, M.; Gessner, U.: Detection of hedgerows in Bavaria, Germany using orthophotos and deep learning. Submitted to Remote Sensing Applications: Society and Environment

For further information please contact: ursula.gessner@dlr.de