Urban Tree Cover Dataset (UTCD)
Description
This dataset is composed of co-registered pair of 10m Sentinel-2 image and 2 m urban tree cover label image. Senitnel-2 image has cropped to patches with size of 128*128 pixels, label image has cropped to pathces with size of 640*640 pixels. Totally 1946 patches are obtained. In label images, the urban tree class and non-tree class are annotated, the statistics of which are 0.1321 billion and 0.6650 billion pixels, respectively. In order for training and testing of a deep neural network, the UTCD is preliminarily split into training set, validation set and test set, with patches of 1383, 195, 368 patch pairs, respectively.
Files
Steps to reproduce
The dataset is composed of co-registered pairs of 10m Sentinel-2 images and 2m urban tree cover label images. Sentinel-2 image is downloaded at Google Earth Engine (GEE) platform, the dataset repository is at https://developers.google.com/earth-engine/datasets/catalog/COPERNICUS _S2#description. The download code can be access https://code.earthengine.google.com/176a2a8e2a3db17c7140a9619f182a4e. Urban tree cover label image is obtained by annotation of ~2m Google Earth images by surveying and mapping experts. The Google Earth images is downloaded using Bigemap software (http://www.bigemap.com/) at 16 levels.