|Description (include details on usage, files and paper references)||The proliferation of satellite imagery has given us a radically improved understanding of our planet. It has enabled us to better achieve everything from mobilizing resources during disasters to monitoring effects of global warming. What is often taken for granted is that advancements such as these have relied on labeling features of significance like building footprints and roadways fully by hand or through imperfect semi-automated methods.
As these large, complex datasets continue to increase exponentially in number, the Defence Science and Technology Laboratory (Dstl) is seeking novel solutions to alleviate the burden on their image analysts. In this competition, Kagglers are challenged to accurately classify features in overhead imagery. Automating feature labeling will not only help Dstl make smart decisions more quickly around the defense and security of the UK, but also bring innovation to computer vision methodologies applied to satellite imagery.
this competition, Dstl provides you with 1km x 1km satellite images in both 3-band and 16-band formats. Your goal is to detect and classify the types of objects found in these regions.
3- and 16-bands images
There are two types of imagery spectral content provided in this competition. The 3-band images are the traditional RGB natural color images. The 16-band images contain spectral information by capturing wider wavelength channels. This multi-band imagery is taken from the multispectral (400 – 1040nm) and short-wave infrared (SWIR) (1195-2365nm) range. All images are in GeoTiff format and might require GeoTiff viewers (such as QGIS) to view. Please refer to our tutorial on how to programmatically view the images.
Ortho Ready Standard Imagery © 2016 DigitalGlobe, Inc.
All imagery credit to: Satellite Imagery © DigitalGlobe, Inc.
Sensor : WorldView 3
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 :
Multispectral: 1.24 m
SWIR: Delivered at 7.5m
Panchromatic and multispectral : 11-bits per pixel
SWIR : 14-bits per pixel
In a satellite image, you will find lots of different objects like roads, buildings, vehicles, farms, trees, water ways, etc. Dstl has labeled 10 different classes:
Buildings - large building, residential, non-residential, fuel storage facility, fortified building
Misc. Manmade structures
Track - poor/dirt/cart track, footpath/trail
Trees - woodland, hedgerows, groups of trees, standalone trees
Crops - contour ploughing/cropland, grain (wheat) crops, row (potatoes, turnips) crops
Vehicle Large - large vehicle (e.g. lorry, truck,bus), logistics vehicle
Vehicle Small - small vehicle (car, van), motorbike
Every object class is described in the form of Polygons and MultiPolygons, which are simply a list of polygons. We provide two different formats for these shapes: GeoJson and WKT. These are both open source formats for geo-spatial shapes.
Your submission will be in the WKT format.
In this dataset that we provide, we create a set of geo-coordinates that are in the range of x = [0,1] and y = [-1,0]. These coordinates are transformed such that we obscure the location of where the satellite images are taken from. The images are from the same region on Earth.
To utilize these images, we provide the grid coordinates of each image so you know how to scale them and align them with the images in pixels. You need the Xmax and Ymin for each image to do the scaling (provided in our grid_sizes.csv) Please refer to our tutorial on how to programmatically view the images.