DeepForest - Street Trees Dataset

This dataset is for Sup. 3 to reproduce the streets data. See Weinstein et al. 2020 "DeepForest: A Python package for RGB deep learning tree crown delineation". 

The Oregon Street Trees data came from the Oregon Parks and Recreation Department. The following data are attached. After receiving a large RGB tile from the Oregon Imagery Program, the image was cropped to form a small training (street_trees_train.tif) and test area (street_trees_test.tif). The train data was then hand-annotated in QGIS but manually reviewing visible trees in the imagery (train.shp). DeepForest was then used to cut the test tile into 400px tiles and predict individual tree crowns. These crowns were compared to field-collected stems available from the Portland Tree survey (https://gis-pdx.opendata.arcgis.com/datasets/street-trees). These trees were lightly cleaned by eye to remove obvious trees that were outdated (e.g. in the middle of buildings in the present-day imagery).

本数据集用于复现补充材料3中的街道数据。详见Weinstein等人2020年发表的《DeepForest：一款用于RGB深度学习树冠勾画的Python工具包》。

俄勒冈街道树木数据集源自俄勒冈州公园与娱乐管理局（Oregon Parks and Recreation Department）。本次附带如下数据：我们从俄勒冈影像项目（Oregon Imagery Program）获取大型RGB瓦片后，将其裁剪为小型训练区域（street_trees_train.tif）与测试区域（street_trees_test.tif）。随后我们在QGIS中对训练数据开展人工标注工作，具体为通过目视解译影像中的可见树木生成train.shp标注文件。随后使用DeepForest工具将测试瓦片切割为400像素（400px）的子瓦片，并完成单棵树冠的预测任务。将预测得到的树冠与波特兰树木调查项目（Portland Tree Survey）公开的野外实测茎干数据进行比对，该数据集可从https://gis-pdx.opendata.arcgis.com/datasets/street-trees获取。我们对该数据集进行了轻度人工清洗，移除了明显过时的树木记录，例如当前影像中位于建筑物正中央的树木。