uavClassification

Script with functions to classify very high resolution orthoimages aquired for example by drone flights.

Module

The functions in the module are mostly wrappers for OTB functions as well as reading and writing raster images with the rasterio package. Example of a workflow using these functions can be found in script.py or test_smoothing_params.py

App with Qt Gui

The script app_seg_params.py contains a gui application where an image can be selected and different segmentation parameters as well as classification parameter can be tested interactively.

For testing purposes example data can be found in the data directory and is also hard coded as defaults when no data is selected with the corresponding buttons in the app.

Currently there is no strategy implemented to reduce computation time when big images are loaded. Thus when changing the sliders for the segmentation parameters there will possibly be a considerably delay in the display of the result.

Installation

In order to use the module itself it is sufficient to download all files or clone the repository and meet the requirements in the requirements.txt file.

If the app is to be used it is necessary to install Qt5 as well as PyQt5. The displaying of results in the app is done with the help of the pyqtgraph module which can be downloaded here http://www.pyqtgraph.org. Untar the archive and place the pyqtgraph directory in the same directory as the app script.

Known Bugs/ Caveats

In generall this is a work in progress (also see todos). Unfortunatelly while the wrapper functions in this module for OTB are in principle designed for it, the input and ouput to the OTB functions as numpy arrays currently does not work as expected. In consequence some workflows are a little cumbersome due to the fact that intermediate output hast to be written to disk and subsequently be read in again for some steps.

Todo

In no particular order:

• Create a sampling construction for the app when it is used with larger images in order to minimize delay of slider change and display of result
• Add further parameters to wrapper functions (see also the documentation in the functions itself)
• Add functions for estimation of segmentation and classification parameters
• Add further filtering options
• Better handling of big images when using raster output in segmentation
• etc.