0% found this document useful (0 votes)
16 views42 pages

Multi Spectral Image Processing

The document outlines a comprehensive process for analyzing multi-spectral imagery, including steps for image acquisition, processing, and analysis using tools like Agisoft Metashape and QGIS. It details the necessary resources, calibration methods, and formulas for calculating vegetation indices such as NDVI, NDRE, and GCI. Additionally, it emphasizes the importance of archiving data for future machine learning applications and provides insights into visualizing changes over time in crop health and other attributes.
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
16 views42 pages

Multi Spectral Image Processing

The document outlines a comprehensive process for analyzing multi-spectral imagery, including steps for image acquisition, processing, and analysis using tools like Agisoft Metashape and QGIS. It details the necessary resources, calibration methods, and formulas for calculating vegetation indices such as NDVI, NDRE, and GCI. Additionally, it emphasizes the importance of archiving data for future machine learning applications and provides insights into visualizing changes over time in crop health and other attributes.
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd

UAS Multi-Spectral image Analysis

Analysis on Multi-Spectral Imagery for Profiling


THIS ASSUMES IMAGE ACQUISITION
IS DONE
The following are steps AFTER the Multi-Spectral Image Acquisition phase is
completed. Raw 5-Band Photos and the Reflectance Panels are assumed
to be in order and available
Mapping Process

Acquisition Processing Analysis


The acquiring of the images Using purpose built Importing the data from
necessary to produce the hardware/software to create Processing to the local GIS
Orthophoto Map and 3D Point the Orthophoto mosaic and
Clouds. Digital Surface Model Raster Manipulation (VI)

Site preparation Additional process for Digital Analysis of the data


Terrain Model Presentation of results
Ground Resolution 2-5cm/pixel
Exporting and converting the Data archive, access, and
Flight Plan files for GIS applications retrieval
Ground Control (if necessary)
100-200 ha per team per day
Mica Sense Red Edge MX Raw Bandwidth

 Band 1 = BLUE; Center wavelength is 475nm width is 32nm


 Band 2 = GREEN; Center wavelength is 560nm width is 27nm
 Band 3 = RED Center wavelength is 668nm width is 14nm
 Band 5 = RED EDGE; Center wavelength is 717nm width is 12nm
 Band 4 = NEAR INFRA RED; Center wavelength is 842nm width is 57nm
NECESSARY RESOURCES

 GIS Workstation with Licensed Agisoft Metashape


 Raw Multi-Spectral Data Set from Drone flight with Reflectance Panel
Photos
 CSV file of the Calibrated Reflectance Panel (email serial number of the
panels to support@[Link] to receive this file) You may also
manually encode the panel values on the provided fields.
 QGIS – for Vegetation Index Visualization and Band Histogram Export
 Google Earth Pro –for KMZ Visualization with Georeferencing
Processing Multi-Spectral Images in
Agisoft Metashape
Process and export a 5-band NORMALIZED (0-1) GeoTIFF
Export Digital Elevation Model (DEM)
Option to Export Digital Terrain Model (DTM)
Preparing the Photos for Alignment

 Open Agisoft Metashape: Workflow: Add Folder: Select 1st Folder


 Select Multi-Camera System: Confirm Calibration Images moved to
separate folder
 Repeat for additional folders until all photos are uploaded
 Select any photo: right click: Estimate Image Quality: All Cameras
 After Image Quality Estimation: Click Photo Details and Sort from Lowest to
highest
 Remove photos below 0.5 quality
Reflectance Panel Calibration

 Tools: Calibrate Reflectance: Locate Panels


 NOTE Ensure you have registered your Panels with MicaSense by emailing
the reflectance panel serial number to support@[Link]. If not you
may encode the panel calibration numbers manually in the provided
space.
 “Use Reflectance panel” should be checked then also Check “Use Sun
Sensor” then OK
Aligning and Optimizing the Photos

 Workflow: Align Photos: Note setting on


screenshot
 After the Alignment Process; Select the
Reference Tab
 Then Click Optimize Cameras with the
settings as shown in the screen shot
 Save the Project with your file naming
and archiving convention
Dense Cloud and Classifying Ground
Points
 Workflow: Build Dense Cloud
 Set Quality and Depth Filtering
 Check “Calculate Point Colors” and run
the process.
 After Dense Cloud Process you can now
view your 3D Point Cloud. Now we classify
ground points
 Tools: Dense Cloud: Classify ground Points
(see screenshot for settings)
 Save your project
DEM, Orthophoto, and Boundaries

 At this point your point cloud will be dark.


This is NORMAL. Improve visual by Tools:
Set Brightness: 500; adjust as needed
 Workflow: Build DEM
 Workflow: Build Orthomosaic
 Draw Polygon: Set Boundary Type: Outer
Boundary (red-black line)
 Export DEM / DTM (select Ground Points
Only)
 Save Project
Raster Transformation and GeoTIFF
Export
 We will now Transform (Normalize) our
Raster file and Export it to a 5-Band
GeoTIFF
 Select Workspace Tab: Orthomosaic: Set
Raster Transform
 Enter the formula B1/32768…B5/32768
 Save Project
 Export Orthophoto: Set Raster Transfor=
INDEX VALUE
 Save file with your Naming Convention
Data Sets Produced

 5-Band GeoTIFF File


 DEM GeoTIFF File
 DTM GeoTIFF (Option) File

 We Can Now use QGIS to Render Visualizations and Histograms from the
Data Above
NOTE: Agisoft Exported GeoTIFF Band Order

 Band 1 = BLUE; Center wavelength is 475nm width is 32nm


 Band 2 = GREEN; Center wavelength is 560nm width is 27nm
 Band 3 = RED Center wavelength is 668nm width is 14nm
 Band 4 = RED EDGE; Center wavelength is 717nm width is 12nm
 Band 5 = NEAR INFRA RED; Center wavelength is 842nm width is 57nm
Creating VI Renders and Histograms
from 5-Band GeoTIFF using QGIS
We use QGIS to export VI Renders and Histogram Values for Profiling
VI FORMULAS to be Used

 NDVI = (R-NIR)/(R+NIR)
 NDRE = (RE-NIR)/(RE+NIR)
 GCI = (NIR/G)-1
PROCESS WALKTHROUGH
Loading Files and Set Layer Parameters
 Open QGIS and drag the GeoTIFF Files
from Agisoft into the workspace
 Select the Normalized 5-Band GeoTIFF
 Ensure you have the Layer Panel Styling
Panel Open
 Set Red Band to Band 3
 Set Blue Band to Band 1
 Set Min/Max Value CumCount@ 0.2 – 99.8
 Adjust Brightness and Contrast
 Select Histogram: Save Plot
 Save Project File
Normalized Difference Vegetation
Index (NDVI)
(Near Infra red – Red)/(Near Infra Red + Red)
(Band5 – Band3) / (Band5 + Band3)
NDVI Visualization and Histogram
 Normalized Difference Vegetation Index will show Vegetation,
Water, and Soil separation and is useful in identifying where
there is dense and no vegetation
 Select Raster: Raster Calculator;
 Enter the Output Filename; Enter the formula (NIR-R)/(NIR+R) or
(Band5-Band3)/(Band5+Band3)
 This will create a new layer with the filename – select this layer
 Select Single band Pseudocolor in Layer Style; Min Max
Cum/Count is 0.2 to 99.8; Linear Interpolation; and Load the
NDVI Color Ramp (create new color ramp: cpt-city: Qgis: Grass:
NDVI) mode=Continuous; Classes 21
 Export the Layer as a Rendered GeoTIFF
 Export The Histogram for profiling
 Shift between the RGB image and NDVI to profile the
reflectance values and note the characteristics.
NDVI Visualization and Histogram
Normalized Difference Red Edge
(NDRE)
(Near Infra red – Red Edge)/(Near Infra Red + Red Edge)
(Band5 – Band 4) / (Band5 + Band4)
NDRE Visualization and Histogram
 Normalized Difference Red Edge is useful for segregating
spectral profiles as the Red Edge is a much narrow band than
Red. Select Raster: Raster Calculator;
 Enter the Output Filename; Enter the formula (NIR-RE)/(NIR+RE)
or (Band5-Band4)/(Band5+Band4)
 This will create a new layer with the filename – select this layer
 Select Single band Pseudocolor in Layer Style; Min Max
Cum/Count is 0.2 to 99.8; Linear Interpolation; and Load the
RdYlGn Color Ramp mode=Continuous; Classes 5
 Export the Layer as a Rendered GeoTIFF
 Export The Histogram for profiling
 Shift between the RGB image and NDRE to profile the
reflectance values and note the characteristics.
NDRE Visualization and Histogram
Green Chloropyll Index (GCI)
(Near Infra red / Green) - 1
(Band5 / Band 2) - 1
GCI Visualization and Histogram
 You can use Green Chlorophyll Index (GCI) to
determine relative plant health in the area. Select
raster; Raster Calculator
 Enter the Output Filename; Enter the formula
(NIR/G)-1 or (Band5/Band2)-1
 This will create a new layer with the filename –
select this layer
 Select Single band Pseudocolor in Layer Style; Min
Max Cum/Count is 0.2 to 99.8; Linear Interpolation;
and Load the RdYlGn Color Ramp
mode=Continuous; Classes 5
 Export the Layer as a Rendered GeoTIFF
 Export The Histogram for profiling
 Shift between the RGB image and GCI to profile the
reflectance values and note the characteristics.
NDRE Visualization and Histogram
Change Over Time
After a regular archive of processed VI Images, you can overlay these
in a GIS software to visualize the changes.
There are also other online services that have computer learning
software for identifying specific image attributes over a large
database (such as “[Link]”)
Sample Reports
NDVI for Vegetation Monitoring, Missing Hills and Plant Count
Missing Hills Located
NDRE for Weed and Water Pooling Detection
Weeds and Water Pooling Located
GCM for Plant Health Identification
The map above shows Chlorophyll levels of your crops which indicate plant health
MOSAIC DETECTED and Located
Framework / Roadmap
Framework/Roadmap

•Acquire Data Per Area Regularly for Monitoring, Validation and Reporting (Multi-Spectral Drone)
•Daily Acquisition of controlled areas (Multi-Spectral Drone) for additional baselines.
ACQUIRE

•Use Agisoft Metashape to Produce Per Plot Data Sets Comprised of: 1. DEM; 2. 5-Band GeoTIFF
•Use QGIS to produce Rendered GeoTIFF and Plot Histogram Values for Profiling
PROCESS •Archive the data sets for future machine learning and refining the profiles

•Load the Rendered GeoTIFF into Machine Learning Platform for scaled analysis
•Use AI, or other capable software, to run the algorithms and determine the issues over the data.
•Create and refine the profiles and update machine learning for faster interpretation
INTERPRET

•Create a granular report which can be distributed via cloud onto a tablet that has GPS and can load the
produced map for field reference.
•Other data goes into a cloud based GIS for archive and management access.
REPORT •PDF Version may be printed as needed from the interface as a “PDF export” button
Sample Data Set Using Normalized
Difference Vegetation Index (NDVI)

Soil@ 0.15; Weeds@ 0.8; Crops@ 0.7


NOTE: Traditional NDVI will not
separate the weeds as well as the
other indexes
Sample Data Set Using Normalized
Difference Red Edge Index (NDRE)

Soil@ 0.05; Weeds@ 0.25; Crops@ 0.15


NOTE: Modern NDRE has better
separation but there are many others
we must try!!!
Analysis on CROP ISSUES using Histogram Profiles
By dividing the plot into “GRIDS” and identifying “GOOD GRIDS” vs “ISSUES GRIDS” we are looking at the parameters
that would differ to later train the computer to look for these “anomalies”
On-Line Cloud Based Services

 Machine Learning on image recognition [Link]


 Orthophoto analysis [Link]
 Cloud based archive and sharing platform [Link]
 Stand alone on-premise/cloud analysis software
[Link]
 Enterprise GIS [Link]
Thank You
STAY SAFE and
GOD BLESS US ALL

You might also like