"Weed Mapping in Maize Fields Using Object-Based Analysis of UAV Images"

Cool article from the open access scientific journal PLOS One

Abstract

The use of remote imagery captured by unmanned aerial vehicles (UAV) has tremendous potential for designing detailed site-specific weed control treatments in early post-emergence, which have not possible previously with conventional airborne or satellite images. A robust and entirely automatic object-based image analysis (OBIA) procedure was developed on a series of UAV images using a six-band multispectral camera (visible and near-infrared range) with the ultimate objective of generating a weed map in an experimental maize field in Spain. The OBIA procedure combines several contextual, hierarchical and object-based features and consists of three consecutive phases: 1) classification of crop rows by application of a dynamic and auto-adaptive classification approach, 2) discrimination of crops and weeds on the basis of their relative positions with reference to the crop rows, and 3) generation of a weed infestation map in a grid structure. The estimation of weed coverage from the image analysis yielded satisfactory results. The relationship of estimated versus observed weed densities had a coefficient of determination of r2=0.89 and a root mean square error of 0.02. A map of three categories of weed coverage was produced with 86% of overall accuracy. In the experimental field, the area free of weeds was 23%, and the area with low weed coverage (<5% weeds) was 47%, which indicated a high potential for reducing herbicide application or other weed operations. The OBIA procedure computes multiple data and statistics derived from the classification outputs, which permits calculation of herbicide requirements and estimation of the overall cost of weed management operations in advance.

Read the resthere

Views: 3458


Moderator
Comment by Gary Mortimer on October 17, 2013 at 3:41am

@Koen gary@suasnews.com yes please ;-) I have stacks of suitable images to try.

Comment by koen.hufkens on October 17, 2013 at 3:58am

@Dries, sadly I won't be able to make it... I'm moving to Harvard to take up a new research position next week.

@Gary, I'll put together two toolboxes (and put them on my website). I have one for R (which is rather slow on large images > 12 Mpx) but works fine for anything smaller. For larger images I rewrote the code in Matlab. Sadly Matlab isn't open source, so you might not have access to a license. I'm working on porting it to Octave as the latest version includes all the image processing routines I need.


Moderator
Comment by Gary Mortimer on October 17, 2013 at 4:13am

Brilliant Koen, thanks.

Comment by koen.hufkens on October 17, 2013 at 5:20am

@Gary, all,

Some literature / methodology concerning the radially averaged FFT spectrum based (FOTO) approach:www.khufkens.com/software/FOTO_literature.zip

The R FOTO toolbox:

www.khufkens.com/software/FOTO_R_toolbox.zip

The matlab FOTO toolbox:

www.khufkens.com/software/FOTO_matlab_toolbox.zip

The use is pretty straightforward, unzip the toolbox set the directory containing the files as working directory, load your image and run the routine.

The matlab routine is the fastest but requires a license, the R code is open source and free. A GNU Octave version of the code is coming soon (I'm still on Ubuntu 12.04 and the latest release which is required for blockprocessing is not in the default repository).

The method is implemented as described in the paper by Proisy et al. Post processing of the r-spectra using partitioned normalization (to account for differences in acquisition circumstances) has to be implemented still.

I haven't had time yet to set up a github repository for some of my code but this will happen in the near future. Given the lack of a version system and it not being public until now this is still pretty dirty code (no error trapping). Bug fixes etc are always welcome...

Cheers,

K

Comment by Ned Horning on October 17, 2013 at 5:22am

There is a lot of information in texture data but it works even better if it's combined with spectral data and other layers (predictor variables) that might be derived from the spectral data or from other sources such as a DEM.


If anyone is interested in trying the object-based approach there are some open source packages that allow you to do that. I'm working on a guide that uses R and the Orfeo Toolbox's segmentation application. The R scrips and a guide to use them are on a bitBucket site: https://bitbucket.org/rsbiodiv/. This is a work in progress but I'm looking for feedback from people interested in using it. It was designed with satellite image classification in mind but I'm also using it for low altitude point-and-shoot images.


On a related note I also work on software that processes near-IR imagery form a single camera like the PLOTS infragram cameras (http://publiclab.org/wiki/near-infrared-camera) or dual visible/NIR cameras. The software is available as an ImageJ/Fiji photo-monitoring plugin on my Github site: https://github.com/nedhorning/PhotoMonitoringPlugin. As with the R scripts I”m looking for people to provide feedback.


I'm not a very good programmer but am looking for ways to make the software more accessible. I would really enjoy working with other people on this. Koen, I'm in Vermont, not too far from Harvard so maybe we can chat some time. I think we have a pretty good handle on acquiring imagery but efficiently extracting useful information is much more difficult. I think a hybrid manual/automated interpretation/classification approach is the biggest bang.

Comment by koen.hufkens on October 17, 2013 at 6:36am

@Ned

I'll be in US as of next week Friday. I'm involved in tracking phenology with conventional cameras (a related topic in a way). The lab has also been experimenting with infragram based cameras, this looks promising (as it's cheaper than any hyperspectral or multispectral imager on the market.

I'll have look at your code. I've been playing with the Orfeo toolbox using stereo pairs to back out DTMs. So I'm familiar with it.

Comment by koen.hufkens on October 17, 2013 at 6:43am

@Ned, for your co-registration (to boost the number of channels) do you know this supplier of z-brackets (http://www.digi-dat.de/produkte/index_eng.html#zbarcanon). Together with a normal and modified camera this would provide a really compact package.

Comment by Ned Horning on October 17, 2013 at 6:56am

@Koen, I'm not familiar with that site. I usually make my own camera mounts out of sheet aluminum but they aren't very nice. Are you working with the PhenoCam folks?

Comment by koen.hufkens on October 17, 2013 at 7:01am

@Ned, yes I'm working with the PhenoCam folks. I was at BU a few years ago more on the satellite remote sensing side of it but heavily involved in the camera based work. The GUI as provided by the PhenoCam website is my work. Together with Oliver a colleague I also established the de facto protocols to be followed.

Comment by Charles Radley on December 15, 2013 at 3:12pm

Is there anything happening to commercialize this technology?  e.g. offering  real time image analysis services for farmers ?  I would be interesting in the  current status, thanks.

Comment

You need to be a member of DIY Drones to add comments!

Join DIY Drones

Groups

Season Two of the Trust Time Trial (T3) Contest 
A list of all T3 contests is here. The current round, the Vertical Horizontal one, is here

© 2020   Created by Chris Anderson.   Powered by

Badges  |  Report an Issue  |  Terms of Service