Skip to content

Collection of tasks for preprocessing multi-sensor multispectral UAV imagery

License

Notifications You must be signed in to change notification settings

envi-idl/UAVToolkit

Repository files navigation

UAVToolkit

This collection of code is meant for preprocessing drone data that has band-band alignment issues. Specifically, it was designed to work with the MicaSense RedEdge sensor in order to process the data and make it ready for ingestion into image stitching software packages. The benefit of processing on your own for the RedEdge sensor is that you don't need to pay monthly fees to process your data on the Atlas platform.

Over time these tools have grown into more than just processing for the RedEdge sensor. There is now an option for working with generic sensor formats where each band comes as a separate image format. Additional support has also been added for the Parrot Sequoia sensor which is another common, inexpensive, multispectral drone sensor.

If you have downloaded this repository, then the best way to view the documentation is to open up index.html in the docs folder.

Requirements

ENVI 5.5 and IDL 8.7.

Older versions of ENVI + IDL are not supported because of the ENVI Task schema definitions that are used. If you are on an older version and there are issues, then you will need to upgrade to the latest version of ENVI + IDL.

Changes

See CHANGELOG.md for complete details on what is new in the latest version.

Worth Mentioning

If there are any feature requests, don't hesitate to add an issue for the repository. If you have an issue, please be sure to include debug information and you may be asked to share some sample data for testing purposes which will not be shared.

Installation

Please select one of the below options. Why one? Well, if you have SAVE files and PRO code on IDL's search path (or installed under ENVI), then the SAVE file will win over the PRO code. This means that you can have fun debugging odd issues with conflicts between newer and older versions of code.

If upgrading, make sure you remove any previous version of the UAV Toolkit from IDL's search path or, ENVI's custom_code, or ENVI's extension folder.

Basic - PRO code

To install the code, you just need to add this repository to IDL's search path. You can access this setting in Preferences -> IDL -> Paths -> Insert, navigate to the directory and select OK, and then make sure to check the box to have this folder recursively searched for code. Click Apply and OK once the path has been added.

Note that you may have to reset IDL's session for the changes to take place. To do this you can simply press the Reset button in the IDL workbench.

ENVI Extensions - SAVE files

For users that want to use the UAV Toolkit as an ENVI extension, you will need to run the build_uav_toolkit.pro file in the primary directory of the repository. Once this completes, take the entire UAVToolkit-build folder and place it into ENVI's extensions folder. Here are the potential locations that you

  • For ENVI 5.5 and admin rights the directory for Windows is C:\Program Files\Harris\ENVI55\extensions

  • If you do not have admin rights then you can find the local user directory in ENVI's preferences under:

    File -> Preferences -> Directories -> Extensions Directory

    Once you place the file in ENVI's extension folder you must restart ENVI before you will see them appear.

Note the location that you installed the SAVE files to. If you get a later version of this code, then you will need to update or remove these files for access to new features.

Note: if you are an experienced ENVI + IDL user, you may be wondering why the instructions are to place the ENVI Tasks in the extensions folder (they usually go in the custom_code directory). The reason that this works is because, when the buttons are added to ENVI, I search for and open all tasks in the extensions_folder meaning that you only have to place the UAV toolkit in one location instead of two which makes the installation process easier.

Usage

To take advantage of the UAV toolkit, you just need to add the source code for this repository to IDL's search path and include the following in and routines that you want to use the band alignment tasks for:

;set IDL's compile options
compile_opt idl2

;start ENVI - can be headless or GUI
e = envi()

;run the UAV Toolkit routine
uav_toolkit

Including idl uav_toolkit ensures that the tasks that ship as a part of this repository are found by ENVI without needing to install them in ENVI's custom code directory.

Below you can find several examples for how you can use the different tasks in the UAV Toolkit to process different UAV multispectral datasets, but first a few notes on the tools.

Running the Examples Below

To run the examples below, the easiest method is to create a new PRO file in the IDL Workbench. To do this, simply:

  1. Click on the New Pro icon the the workbench

  2. Copy the example code to the window

  3. Add and end statement as the last line in the file

  4. Save the code with a .pro file extension

  5. Press Compile and then Run to execute the code.

Notes

  • Calibration:

    If you have reflectance panels, then these images will need to be separated from the main flight images. To work with the UAV Toolkit, you will need to place all reflectance panels in a folder called "reflectance_panels" in your data directory. This makes sure that these scenes can't be chosen for generating reference tie points. To have the reflectance panels be applied to convert the scenes to reflectance, then you will need to set the CO_CALIBRATION parameters to for the UAVBandAlignment task. This will correct for any differences in exposure time, ISO sensitivity to light, and perform the calibration step if the panels are present.

    When using reflectance panels, you will also need to specify the percent reflectance for the panel surface (i.e. 70%). To do this, you can use the PANEL_REFLECTANCE task parameter to specify a 0 to 100 floating point value for this.

  • The algorithm may have a harder time finding reference tie points for the Parrot Sequoia because the multispectral sensor has a fisheye lens, which makes the geometry more complicated. The tools shown below have worked great on sample datasets, but that may not apply to all collections. It should, but it is just worth setting expectations a bit.

Generic Sensor

In order to process generic sensors with the UAV Toolkit, you will need datasets that consist of multipage TIFF or contain a single band per separate image file. Each band should have:

  1. Similar data range and data type.

  2. Same image dimensions

If these two conditions are met, then you just need to have the unique file identifiers that can be used to group the separate images into a single raster. Here is an example of an image group and the file identifiers for a dataset where each band is a separate file:

  • img_001_1.tif

  • img_001_2.tif

  • img_001_3.tif

  • img_001_4.tif

And the file identifiers look like:

  • ['_1.tif', '_2.tif', '_3.tif', '_4.tif']

This groups the images above into a single image. If we want to then generate and apply reference tie points for this generic group then we can simply do the following in IDL:

;set IDL's compile options
compile_opt idl2

;start ENVI
e = envi(/HEADLESS)

;get UAV Toolkit ready
uav_toolkit

;set up task
alignTask = ENVITask('UAVBandAlignment')
alignTask.SENSOR = 'generic'
alignTask.FILE_IDENTIFIERS = ['_1.tif', '_2.tif', '_3.tif', '_4.tif']
alignTask.INPUTDIR = 'C:\data\directory'
alignTask.GENERATE_REFERENCE_TIEPOINTS = 1
alignTask.APPLY_REFERENCE_TIEPOINTS = 1
alignTask.execute

Multi-page TIFFs

If your dataset consists of multi-page TIFFs, then you can just specify a single file-identifier as below to process the data:

;set IDL's compile options
compile_opt idl2

;start ENVI
e = envi(/HEADLESS)

;get UAV Toolkit ready
uav_toolkit

;set up task
alignTask = ENVITask('UAVBandAlignment')
alignTask.SENSOR = 'generic'
alignTask.FILE_IDENTIFIERS = ['.tif']
alignTask.INPUTDIR = 'C:\data\directory'
alignTask.GENERATE_REFERENCE_TIEPOINTS = 1
alignTask.APPLY_REFERENCE_TIEPOINTS = 1
alignTask.execute

MicaSense RedEdge

A custom batch processing routine has been developed for the MicaSense RedEdge sensor to make processing the data very simple. To process the data, you can simply do the following using IDL:

;set IDL's compile options
compile_opt idl2

;start ENVI headlessly
e = envi(/HEADLESS)

;initialize the uav_toolkit
uav_toolkit

;this folder should be the one that contains the unzipped contents of the sample data.
flightDir = 'C:\data\rededge'

;create our batch task
redEdgeTask = ENVITask('UAVBatchRedEdge')
redEdgeTask.FLIGHTDIR = flightDir
redEdgeTask.execute

Where FLIGHTDIR is the folder that contains the RedEdge data. In other words, FLIGHTDIR should contain the folders 000 and 001 as collected by the sensor. These folders must all be from the same sensor and be at approximately the same height off of the ground for the reference tie points to be valid.

If you have reflectance panel images in each data folder, then create a subdirectory called "reflectance_panels" and place them in there. These files will be excluded from processing and only used to calibrate the data.

You can also use the ENVITask UAVBandAlignment directly to process a single folder of data using:

;set IDL's compile options
compile_opt idl2

;start ENVI
e = envi(/HEADLESS)

;get UAV Toolkit ready
uav_toolkit

;set up task
alignTask = ENVITask('UAVBandAlignment')
alignTask.SENSOR = 'rededge'
alignTask.INPUTDIR = 'C:\data\directory'
alignTask.GENERATE_REFERENCE_TIEPOINTS = 1
alignTask.APPLY_REFERENCE_TIEPOINTS = 1
alignTask.execute

Customizing Batch RedEdge Parameters

If you are running the UAVBatchRedEdge task, then you can pass in a custom UAVBandAlignment task to edit all parameters of the registration process. Here is an example of how you can do this to have the UAVBatchRedEdge generate multi-channel TIFFs instead of multi-page TIFFs.

;set IDL's compile options
compile_opt idl2

;start ENVI headlessly
e = envi(/HEADLESS)

;initialize the uav_toolkit
uav_toolkit

;create band alignment task to customize settings
alignTask = ENVITask('UAVBandAlignment')
alignTask.MULTI_CHANNEL = 1

; with custom parameters, we have to tell the UAV Toolkit what to process
alignTask.GENERATE_REFERENCE_TIEPOINTS = 1 ; generate the reference tiepoints to register the images
alignTask.APPLY_REFERENCE_TIEPOINTS = 1    ; register the bands of images we found
alignTask.RIGOROUS_ALIGNMENT = 1           ; slower, but always works

;this folder should be the one that contains the unzipped contents of the sample data.
flightDir = 'C:\data\rededge'

;create our batch task
redEdgeTask = ENVITask('UAVBatchRedEdge')
redEdgeTask.FLIGHTDIR = flightDir
redEdgeTask.BAND_ALIGNMENT_TASK = alignTask
redEdgeTask.execute

Parrot Sequoia

You can also process the Parrot Sequoia data with the UAV toolkit. Note that generating the tie points for this data may take longer because the multispectral bands are taken with a fisheye lens. To process the Sequoia data, just use the UAVBandAlignment task with:

;set IDL's compile options
compile_opt idl2

;start ENVI
e = envi(/HEADLESS)

;get UAV Toolkit ready
uav_toolkit

;set up task
alignTask = ENVITask('UAVBandAlignment')
alignTask.SENSOR = 'sequoia'
alignTask.INPUTDIR = 'C:\data\directory'
alignTask.GENERATE_REFERENCE_TIEPOINTS = 1
alignTask.APPLY_REFERENCE_TIEPOINTS = 1
alignTask.execute

Custom Percent Reflectance

If you know the percent reflectance (from 0 to 100) of your reflectance panel for each of your bands, you can specify them individually with the PANEL_REFLECTANCE task parameter for the tasks demonstrated above. Here is an example with the UAVBatchRedEdge task:

;set IDL's compile options
compile_opt idl2

;start ENVI headlessly
e = envi(/HEADLESS)

;initialize the uav_toolkit
uav_toolkit

;this folder should be the one that contains the unzipped contents of the sample data.
flightDir = 'C:\data\rededge'

;create our batch task
redEdgeTask = ENVITask('UAVBatchRedEdge')
redEdgeTask.FLIGHTDIR = flightDir
redEdgeTask.PANEL_REFLECTANCE = [67, 69, 68, 67, 61] ;must be 0 to 100, NOT 0 to 1.0
redEdgeTask.execute

File Formats

By default the UAV Toolkit generates multi-page TIFFs for the output files. If you want to generate multi-band TIFFs instead, you can do the following for the BandAlignmentTask:

;set IDL's compile options
compile_opt idl2

;start ENVI
e = envi(/HEADLESS)

;get UAV Toolkit ready
uav_toolkit

;set up task
alignTask = ENVITask('UAVBandAlignment')
alignTask.SENSOR = 'sequoia'
alignTask.INPUTDIR = 'C:\data\directory'
alignTask.GENERATE_REFERENCE_TIEPOINTS = 1
alignTask.APPLY_REFERENCE_TIEPOINTS = 1
alignTask.MULTI_CHANNEL = 1
alignTask.execute

Default Values

For each one of the sensors, there are some default task parameters for the UAVBandAlignment task which are custom to each sensor type. Each of these parameters can be overridden by specifying the accompanying parameter in the UAVBandAlignment task object.

These defaults are taken directly from the source code and are:

;==============================================================================================
;check the sensor type and set defaults for our processing
case (parameters.SENSOR) of
  'rededge':begin
    parameters['BASE_BAND'] = 2
    parameters['FILE_IDENTIFIERS'] = ['_1.tif', '_2.tif', '_3.tif', '_5.tif', '_4.tif']

    ;specify the metadata for our output scene
    meta = ENVIRasterMetadata()
    meta['band names'] = ['Blue', 'Green', 'Red', 'Red Edge', 'NIR']
    meta['wavelength units'] = 'nm'
    meta['wavelength'] = [450, 550, 650, 750, 850]
    parameters['METADATA'] = meta

    ;check for other parameter values
    if ~parameters.hasKey('SEARCH_WINDOW_FROM_HEIGHT') then parameters['SEARCH_WINDOW_FROM_HEIGHT'] = 1
    if ~parameters.hasKey('CO_CALIBRATION') then parameters['CO_CALIBRATION'] = 1
    if ~parameters.hasKey('MAX_VALUE_DIVISOR') then parameters['MAX_VALUE_DIVISOR'] = 16s
    if ~parameters.hasKey('MAX_PIXEL_VALUE') then parameters['MAX_PIXEL_VALUE'] = 4095
  end
  'sequoia':begin
    parameters['BASE_BAND'] = 1
    parameters['FILE_IDENTIFIERS'] = ['_GRE.TIF', '_RED.TIF', '_REG.TIF', '_NIR.TIF']

    ;specify the metadata for our output scene
    meta = ENVIRasterMetadata()
    meta['band names'] = ['Green', 'Red', 'Red Edge', 'NIR']
    meta['wavelength units'] = 'nm'
    meta['wavelength'] = [550, 650, 750, 850]
    parameters['METADATA'] = meta

    ;check for other parameter values
    if ~parameters.hasKey('SEARCH_WINDOW_FROM_HEIGHT') then parameters['SEARCH_WINDOW_FROM_HEIGHT'] = 0
    if ~parameters.hasKey('CO_CALIBRATION') then parameters['CO_CALIBRATION'] = 0
    requested_number_of_tiepoints = 500
    minimum_filtered_tiepoints = 100
  end
  else:begin
    ;make sure we have information about our panel identifiers
    if ~parameters.hasKey('FILE_IDENTIFIERS') then begin
    message, 'FILE_IDENTIFIERS not specified in parameters, required for generic sensor!'
    endif

    ;set default parameters
    if ~parameters.hasKey('BASE_BAND') then paramers['BASE_BAND'] = 0
    if ~parameters.hasKey('SEARCH_WINDOW_FROM_HEIGHT') then parameters['SEARCH_WINDOW_FROM_HEIGHT'] = 0
    if ~parameters.hasKey('CO_CALIBRATION') then parameters['CO_CALIBRATION'] = 0
  end
endcase

License

© 2018 Harris Geospatial Solutions, Inc.

Licensed under MIT. See LICENSE.txt for additional details and information.