Tutorial 1: classifying a custom image stack with existing reference data
The following scenario shows the classification workflow based on provided demo data of the Kyritz-Ruppiner Heide. The aim is to find results for 7 classes for a former military training area north east of Berlin, Germany.
Step 0: Creation of a workspace
We start with creating a new workspace by clicking on the Create New
button in step 0. The name is required,
description is optional:
After clicking on Create
we can now proceed with step 1 by clicking on Upload data
.
Step 1: Input of reference data
We can here choose if we want to upload a Shapefile or a table. In this case we choose the option
for the .zip
format and upload our local file.
Option A) Shapefile
This reference data includes a 7 class point shapefile identified by expert knowledge. The classes represent dominant ground cover classes within the Kyritz-Ruppiner Heide and include: deciduous, coniferous, heath young, heath old, heath shrub, bare ground and xeric grass.
The following dialogue allows to select the desired classes. In our example, we keep all classes.
After uploading, the reference points will be displayed on the map and step 2 is now available. By
clicking on Add data
we can proceed with the time series.
Step 2: Input of an image layer stack
Here we have two options: We can upload a local file or use the build-in downloader. We choose
option 1 by clicking on Upload local image layer stack
:
Option A) Upload local image layer stack
When uploading a local image stack we need to provide additional information about the data in a json file
as described in
the documentation.
We click on the first Add data
option and add our json file.
After doing that we can see a preview of the metadata and upload the data stack by clicking on the second Add data
.
Our Sentinel-2 timeseries stack from 2018 consists of data from 6 days, each including 9 bands (total 54 layers).
The image stack will then be rendered on the map:
Step 3: Surface classification
We are now able to start the classification by clicking on Start classification
. We can now change the parameters
(in this case we will stick to the default settings) and start the classification of the first class by clicking on
Run Classification
.
The classification can take some time, depending on the data size. Attention: each run will provide unique, not reproducible results, although you may follow this tutorial, your results will look different, it is even likely that your first class differs from this example.
In our case, after the first run, the algorithm provides the following result for the class coniferous:
As the result provided enough valid models to continue, we set a suitable threshold, in this example 8 is chosen.
We do this for all the classes. You can also add the legend if you want to see the exact number of models. Click on the lower icon in the upper right corner:
If one is not satisfied with the result and wants to change the parameters, it is possible to click on the Reclassify
button and change the values for the class and run it again. It is always possible to end the classification by clicking on
End classification, save and show results
. If you leave the workspace with an unfinished classification you can continue
later, if you end it by that button it will be considered as finished and can not be continued. After having reached
the last class that button is replaced by Last Class
:
We have now classified all classes and the results are rendered on the map.
Results
In the results we see the distribution of the classes. Pixels that were not classified fall into the class other. A legend can be toggled on and off with the lower button in the upper right corner:
With the upper button in the upper right corner we can now (de)select the layers of the classes:
If we refresh the page we can see our workspace by clicking on Open
in step 0. Here you will have the
options to render, download or delete the workspace: