DroneMapper

We often get requests from our customers to develop additional value-added features and algorithms to extract more meaningful information from aerial or terrestrial data collections. Recently, the area of Precision Agriculture has seen enormous growth triggered by reduced technology costs and overall interest in new/emerging technologies. The integration of UAS systems, GPS, RTK and other geo-spatial technologies into the farming sector allows a whole new world of opportunities. Although technology can't solve all of our tasks, it can certainly help in new and exciting ways! 

A common task for an Agronomist is the creation of management zones or "area's of interest" based on the available data. The data could be an aerial imagery collection in NIR, yield data from a terrestrial collection, EC soil samples, or a combination of all. Historically, a large amount of research has been devoted to automating the process of extracting appropriate zones when high resolution spatial data is available. This can also be a complicated and time consuming task.

"Normalized difference vegetation index (NDVI) are closely related to many vegetation parameters such as leaf area index, vegetation cover, vegetation biomass and crop growth, so is often used to monitor crop growth and predict crop yield." ^[1]

We've developed a set of automated algorithms to quickly exploit the types of data mentioned above to generate zones using imagery processing and computer vision technologies. An example is shown below using the "Precision Agriculture R-G-NIR, Switzerland" example on the following page: https://dronemapper.com/sample_data/

NDVI GeoTIFF

Classified NDVI GeoTIFF

The image above shows a classified NDVI GeoTIFF where each pixel falls into one of the following categories:

• VL – Very Low Health
• L – Low Health
• A – Average Health
• H – Healthy
• VH – Very Healthy
• EH – Extreme Health

Once classification is completed, DroneMapper generates a shapefile with management zones. The shapefile output is UTM WGS84 format and compatiable with major agriculture software providers such as SST, SMS, Apex, etc. 

Classified NDVI GeoTIFF w/ Shapefile Overlay – "A" and "L" Management Zones Highlighted

Shapefile w/ All Management Zones Shown (UTM 32 Northern)

Each polygon in the generated shapefile includes records in the .dbf file with the following information:

• Classification – 'class'
• NDVI Indice Value – 'z'
• Fill Color – 'color'
• Polygon Id – 'id'
• Area Acre – 'area'

This allows sorting or grouping of the polygons based on .dbf column values.

"EH", "A" and "VH" Management Zones w/ NIR Orthomosaic

A 3D Geo-Referenced Crop Health Map (viewed in GlobalMapper)

KML/KMZ Representation of Alerts and Area’s of Interest

The algorithm completes processing after a few minutes and could easily be adapted to provide point data for Crop Scouting locations, ground truthing, etc.

Download the data generated in this post here or on our samples page. For more information on agriculture management zones we recommend the following link. Please let us know if you have any questions!

Here is a simplified Excel spreadsheet that can be used to assist UAS aerial photo mission planning for those that don’t have access to mission planning software either provided with the UAS or custom written. The spreadsheet uses the following inputs (cells are highlighted in yellow) provided by you for your mission:

• Camera and lens parameters including the CCD focal plane size in pixels and millimeters and the actual lens’ focal length. This is the true focal length, not 35 mm equivalent, and should be the value written into the photo’s EXIF metadata during imagery collection. A great source of camera and lens information can be found here: www.dpreview.com. Go to the camera tab at the top of the page and then select camera manufacturer and search for your model. You’ll find focal plane info in the specifications tab.
   
• Mission and imagery collection parameters including UAS ground speed, flight elevation above ground, imagery overlap, both forward and side, and the area of interest (AOI) width and length. The AOI assumes a rectangle with the shorter side the width and longer side the length.

After you have input the various parameters for your mission the spreadsheet provides the following outputs (these cells have no color fill and should be locked to the user):

• Focal plane pixel size,
• Camera shutter speed to minimize image blur,
• Imagery ground sample distance or size of the pixel on the ground,
• Photo or frame size on the ground, width and length,
• Flight line spacing to achieve side lap input and the number of flight lines for the AOI,
• Distance between successive photos in the flight line to achieve forward overlap input and the number of images for each flight line,
• And, the total number of images for the AOI to achieve the coverage desired.

  Suggestions:
   

• Platform ground speed input should consider the fastest speed the UAS could achieve during the mission in case of a tailwind, for example,
• Dronemapper recommends no less than 60% forward overlap and no less than 40% side lap. For difficult homogenous scenes and significant terrain elevation changes one should increase overlaps to 75% forward and 60-75% side.
• Always extend the width and length of the AOI to insure full coverage of the complete scene of interest. This minimizes digital elevation model (DEM) noise at the AOI boundaries.
• Discussion of the image collection geometry and formulas used can be found here: https://dronemapper.com/uas_photogrammetry_processing.

We utilized Globalmapper (GM), version 16.1, to determine stockpile volumes for the BLM gravel pit example. A Canon SX 260 HS was used to collect 240 images at a ground sample distance of 3.2 cm. Dronemapper processed the imagery generating the geo-referenced orthomosaic and DEM for the scene. For this analysis we imported the DEM into GM and performed volumetric estimation of three stockpiles. The three piles are identified in the DEM image below.

After importing the DEM the digitizer tool was used to generate an area at the base of each of the stockpiles. Once the pile areas were defined we applied elevations from the terrain to each of the three pile areas. We then computed pile volumes. GM generates a flattened terrain surface (reference) beneath the pile and a pile surface. The difference between these two 3-D surfaces computes the volume of the pile. The table outlines the measurements made on each pile.

The three measurements are illustrated in the next three screenshots. The last screenshot shows a 3-D view of the identified piles that were measured to give the viewer perception of their relative sizes. Besides cut volumes, GM also performs fill volumes, valuable for the mining and construction industries.

This same technique can be used for estimating above ground biomass for crops. A section of a 3-D model of a research corn field is illustrated below. Imagery was collected at 5 cm GSD in the NIR. One of the plots within the field is identified with a plot shapefile (in yellow). The average plot height was estimated at 216 cm using the volume and area generated by GM. In this case we could not utilize two 3-D surfaces to compute the volume since adjacent plots obscured the terrain below the plot of interest. Instead, we used a fixed mean elevation of the terrain as the reference to compute the volume of the plot’s 3-D surface.

We will be working with precision agricultural specialists this growing season to track a set of crops/fields on a weekly basis. The DEM will be generated for each imagery collection of the crop through the growing season. Biomass volume estimation will be computed for each collection for all of the plots in the field. Plot biomass accumulation will be tracked weekly, statistically analyzed and correlated with plant genetics, field treatments and other phenomenology collections. We hope to report on this by the end of this year – so stay tuned!

We were happily surprised when we learned of the blanket exemption for a flight height of 200 feet or lower, above ground, for commercial UAS operators with the 333 granted. Now, that has sunk in and DroneMapper has had a chance to process a few imagery sets shot at 200 foot – we discuss the implications, both benefits and limitations, from an operational and imagery processing standpoint. First a couple of observations:

• Most UAS that DroneMapper is familiar with are flying compact cameras or in some cases DSLRs. At 200 foot altitude the camera/lens system produces an image on the ground with a nominal pixel size of one inch (ground sample distance/ GSD) or less at the lens’ minimum focal length setting. Without investing in an expensive camera/lens combination that could provide 2 inch GSD or utilizing a fisheye lens (which causes its own issues) you will be imaging at very high resolution.
   
• Scenes that lack structure or features pose difficulties for image-to-image tie point generation and orthomosaicking. As the ground resolution gets finer or smaller this can be a real challenge with scenes like agricultural crop canopies, bare ground, dense forests, water, etc. where each image looks very like the previous or subsequent. Example sequential photos of a field, prior to planting, imaged at one inch GSD illustrates this point.
   
  
• Fully processed, geo-referenced ortho and DEM GeoTiffs are approximately 2-4 GB each for a 250 acre scene imaged at one inch GSD – pretty large files!

Benefits of very high resolution:

• Absolute geo-spatial accuracy of the ortho and DEM utilizing ground control or a RTK GPS solution will be one inch or less horizontal and one to three inches vertical (RMSE), essentially survey grade.
   
• Objects and features sized 3-4 inches can be identified by virtue of the very high resolution.

Limitations/Suggestions – Operational & Processing:

• For very homogeneous scenes, as illustrated above, consider increasing imagery collection overlaps to 75% both in- and cross-track. For a 250 acre field imaged with a 12 Mpixel camera at one inch GSD you’ll need to collect about 2,000 photos. This number can be scaled to your field size, i.e. 125 acres would require about 1,000 photos.
   
• Many UAS/sensor collection platforms will likely be stressed for these scenarios because of sensor triggering frequency or photo time interval, platform ground speed, platform flight endurance and GPS IMU navigational and photo geo-tagging capability.
   
• Processing this number of very high resolution photos for ortho and DEM construction is time consuming and will lead to significantly longer turn-around times for data products.
   
• The size of the ortho and DEM GeoTiffs can present problems for a number of GIS and CAD applications. Download one or more of the example data sets and verify that your application can read the GeoTiff or .ply files properly.
   
• For those that really don’t need very high resolution for their applications DroneMapper will be experimenting with other processing techniques to minimize turn-around time, keep it affordable and deliver the information that you need.

Congratulations to all Exemption 333 holders, the best in your business venture and please contact us if we can help in anyway. The DroneMapper team.

Download the PDF

Original Imagery: @sensefly
Camera: Canon IXUS 127
GSD: 3 cm / px
Images: 98 @ 4608×3456 8bit JPG
SenseFly RTK Report: [download]

GCP: 1 Z_Delta: 0.016 [m] GCP: 2 Z_Delta: 0.027 [m] GCP: 3 Z_Delta: 0.050 [m] GCP: 4 Z_Delta: 0.059 [m] GCP: 5 Z_Delta: 0.060 [m] GCP: 6 Z_Delta: 0.032 [m] GCP: 7 Z_Delta: 0.039 [m] GCP: 8 Z_Delta: 0.021 [m] GCP: 9 Z_Delta: 0.015 [m] GCP: 10 Z_Delta: 0.048 [m] GCP: 11 Z_Delta: 0.034 [m] GCP: 12 Z_Delta: 0.034 [m] GCP: 13 Z_Delta: 0.026 [m] GCP: 14 Z_Delta: -0.010 [m] GCP: 15 Z_Delta: 0.010 [m] GCP: 16 Z_Delta: 0.015 [m] GCP: 17 Z_Delta: 0.038 [m] GCP: 18 Z_Delta: 0.036 [m] GCP: 19 Z_Delta: 0.025 [m]