Urban Tree Cover Dataset (UTCD)

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.本数据集由配准后的10米分辨率哨兵二号（Sentinel-2）影像与2米分辨率城市树木覆盖标签影像组成。哨兵二号影像被裁剪为128×128像素的图像块，标签影像则被裁剪为640×640像素的图像块，最终共得到1946组图像对。标签影像中已标注城市树木类与非树木类两个类别，两类像素的统计总量分别为1.321亿与6.650亿。为支撑深度神经网络的训练与测试工作，该城市树木覆盖数据集（Urban Tree Cover Dataset, UTCD）被初步划分为训练集、验证集与测试集，各组分别包含1383、195与368组图像对。

搜集汇总

背景与挑战

背景概述

UTCD是一个包含Sentinel-2影像和城市树木覆盖标签的配对数据集，用于深度学习网络的训练和测试，包含1946个影像块并已划分为训练、验证和测试集。