多光谱图像数据集

多光谱图像数据集

GitHub

语义分割资源列表
https://github.com/mrgloom/awesome-semantic-segmentation

数据集:

1 dfc2019 track1

https://competitions.codalab.org/competitions/21120#participate

• [02] - Ground
• [05] - High Vegetation / Trees
• [06] - Building Roof
• [09] - Water
• [17] - Elevated Road / Bridge
• [65] - Unlabeled

官方baseline: https://github.com/pubgeo/dfc2019

官方baseline使用基于resnet34的Unet，Unet是网上一个开源的Keras Segment库，官方baseline包含了训练好的权重。

官方baseline使用到以下开源库：

一个数据增强库：albumentations

一个图像分割算法库：segmentation_models

2 AOI_4_Shanghai_Train

SpaceNet 数据集 https://spacenetchallenge.github.io/

8波段图像，0.3m分辨率，有全色图和多光谱图，以及融合后的多波段图像

标注：建筑物区块

├── AOI_4_Shanghai_Train.tar.gz
│      ├── geojson
│      │   └── buildings  # Contains GeoJson labels of buildings for each tile
│      ├── MUL            # Contains Tiles of 8-Band Multi-Spectral raster data from WorldView-3
│      ├── MUL-PanSharpen # Contains Tiles of 8-Band Multi-Spectral raster data pansharpened to 0.3m
│      ├── PAN            # Contains Tiles of Panchromatic raster data from Worldview-3
│      ├── RGB-PanSharpen # Contains Tiles of RGB raster data from Worldview-3
│      └── summaryData    # Contains CSV with pixel based labels for each building in the Tile Set.

3 Dstl Satellite Imagery Feature Detection

Kaggle竞赛
https://www.kaggle.com/c/dstl-satellite-imagery-feature-detection/data

Imagery details

• Sensor : WorldView 3

• Wavebands :

• • Panchromatic: 450-800 nm
  • 8 Multispectral: (red, red edge, coastal, blue, green, yellow, near-IR1 and near-IR2) 400 nm - 1040 nm
  • 8 SWIR: 1195 nm - 2365 nm

• Sensor Resolution (GSD) at Nadir :

• • Panchromatic: 0.31m
  • Multispectral: 1.24 m
  • SWIR: Delivered at 7.5m

• Dynamic Range

• • Panchromatic and multispectral : 11-bits per pixel
  • SWIR : 14-bits per pixel

分类：

1. Buildings - large building, residential, non-residential, fuel storage facility, fortified building
2. Misc. Manmade structures
3. Road
4. Track - poor/dirt/cart track, footpath/trail
5. Trees - woodland, hedgerows, groups of trees, standalone trees
6. Crops - contour ploughing/cropland, grain (wheat) crops, row (potatoes, turnips) crops
7. Waterway
8. Standing water
9. Vehicle Large - large vehicle (e.g. lorry, truck,bus), logistics vehicle
10. Vehicle Small - small vehicle (car, van), motorbike

• sample_submission.csv - a sample submission file in the correct format

• • ImageId - ID of the image
  • ClassType - type of objects (1-10)
  • MultipolygonWKT - the labeled area, which is multipolygon geometry represented in WKT format

4 DeepGlobe

数据集：RGB 语义分割 6+1种类别

Data

• The training data for Land Cover Challenge contains 803 satellite imagery in RGB, size 2448x2448.
• The imagery has 50cm pixel resolution, collected by DigitalGlobe’s satellite.
• You can download the training data in the download page with filetype of “Starting Kit”. Testing satellite images will be will be uploaded later.

Label

• Each satellite image is paired with a mask image for land cover annotation. The mask is a RGB image with 7 classes of labels, using color-coding (R, G, B) as follows.

• • Urban land: 0,255,255 - Man-made, built up areas with human artifacts (can ignore roads for now which is hard to label)
  • Agriculture land: 255,255,0 - Farms, any planned (i.e. regular) plantation, cropland, orchards, vineyards, nurseries, and ornamental horticultural areas; confined feeding operations.
  • Rangeland: 255,0,255 - Any non-forest, non-farm, green land, grass
  • Forest land: 0,255,0 - Any land with x% tree crown density plus clearcuts.
  • Water: 0,0,255 - Rivers, oceans, lakes, wetland, ponds.
  • Barren land: 255,255,255 - Mountain, land, rock, dessert, beach, no vegetation
  • Unknown: 0,0,0 - Clouds and others

• File names for satellite images and the corresponding mask image are _sat.jpg and _mask.png. is a randomized integer.

5 ISPRS：Potsdam，Toronto

由国际摄影测量及遥感探测学会(ISPRS)组织发布

数据集官网：http://www2.isprs.org/commissions/comm3/wg4/2d-sem-label-vaihingen.html

Github上该数据集的一个语义分割开源代码：

https://github.com/nshaud/DeepNetsForEO

5.1 2D Semantic Labeling Contest - Potsdam

The ground sampling distance of both, the TOP and the DSM, is 5 cm. The DSM was generated via dense image matching with Trimble INPHO 5.6 software and Trimble INPHO OrthoVista was used to generate the TOP mosaic. In order to avoid areas without data (“holes”) in the TOP and the DSM, the patches were selected from the central part of the TOP mosaic and none at the boundaries. Remaining (very small) holes in the TOP and the DSM were interpolated.
The TOP come as TIFF files in different channel composistions, where each channel has a spectral resolution of 8bit:

• IRRG: 3 channels (IR-R-G)
• RGB: 3 channels (R-G-B)
• RGBIR: 4 channels (R-G-B-IR)

In this way participants can pick the data needed conveniently.
The DSM are TIFF files with one band; the grey levels (corresponding to the DSM heights) are encoded as 32 bit float values. The TOP and the DSM are defined on the same grid (UTM WGS84). Each tile comes with an affine transformation file (tiff world file) in order to enable a re-composition of images to larger mosaics if desired.

5.2 2D Semantic Labeling - Vaihingen data

The data set contains 33 patches (of different sizes), each consisting of a true orthophoto (TOP) extracted from a larger TOP mosaic, see Figure below and a DSM. For further information about the original input data, please refer to the data description of the object detection and 3d reconstruction benchmark.

http://www2.isprs.org/commissions/comm3/wg4/2d-sem-label-vaihingen.html

6 SAR图像和光学图像

同样由国际摄影测量及遥感探测学会(ISPRS)组织发布。

SEN1-2 Dataset for Deep Learning in SAR-Optical Data Fusion：

http://www2.isprs.org/commissions/comm1/wg3/resources.html

7 TNO Image Fusion Dataset

https://figshare.com/articles/TNO_Image_Fusion_Dataset/1008029

8 RIT-18

https://github.com/rmkemker/RIT-18