3689641838?profile=original

I've recently finished a research project using UAVs to classify land cover types on British upland heath. I'm using an APM controlled hexacopter (H550) carrying a Canon Powershot S95 modified for NIR photography. I've successfully surveyed and classified around 15 hectares in 10 minutes. Using image analysis software I've been able to classify the major land cover types with about 90% accuracy which is comparable to that obtained from imagery captured from much larger platforms using many more spectral bands.

An NDVI colour index of the orthomosaic was created and used in the classification process. This really helps in the spectral differentiation process. I think the results here show massive potential for the frequent, low-cost monitoring of the status of, and temporal changes in (sorry for the academic speak) many habitats, not just heathland.

The images below show the various outputs.

1. The full NIR orthomosaic

Screen%20Shot%202015-03-19%20at%2015.38.15.png

2. The NDVI colour index

Screen%20Shot%202015-03-19%20at%2015.37.50.png

3. The colour index overlaid on the green channel of the orthomosaic highlighting healthy and dead vegetation

Screen%20Shot%202015-03-19%20at%2015.44.42.png

4. The classified land cover types in the GIS software (Purple is Calluna dominated heath, light green is rank grass cover, brown is cut heath and green are trees)

Screen%20Shot%202015-03-19%20at%2015.37.07.png?width=649

I used ImageJ to create the NDVI colour index image, unfortunately it doesn't keep the GeoTIFF data in the output so I've used GDAL to georeference the colour index TIFF

E-mail me when people leave their comments –

You need to be a member of diydrones to add comments!

Join diydrones

Comments

  • Great work Mark, I've been looking for a good example of drone-imagery plant classification and I think this is one.

  • Hi Mark, Which filters did you use in your modified S95 NIR ? and which SW for Orthophoto ? what was the speed /height of the Hex and the rate of taking images, using CHDK or distance based on APM  ?. Thx.

  • Hi James, I don't think that they do in this case. They can have an effect in NDVI analysis alone on hi res images though. As I'm using nearest neighbour analysis with training samples from areas in shadow as well as light shadow effects should be minimised. The survey took place at midday when the sun was highest, this helps to minimise shadow effects.
  • Hi Mark

    Do you think the shadows impact the quality of the image significantly?

  • Moderator

    Its the one to goto they are very nice people and work with the CAA on lots of stuff.

  • I am indeed Gary, I'll take a closer look at Resource.

  • Moderator

    I think you are in the UK Mark, if you do not already have a formal piloting qualification take yourself off to Resource and get an RPQ-S that will make finding work much easier!

  •  thanks for the feedback Koen. I did play with the use of haralick textures in the classification process and it made a big improvement. The problem was with the processing time and power available to me, haralick computation is pretty intensive. I'd love to spend more time on it, there are a huge number of possibilities here. Mac Arthur (2008) had some good results in life stage classification of heather with object based image analysis.

    Sadly my research time is very limited from now :( I'm no longer a student from June and real life beckons. I'm searching for employment opportunities in this field but they're thin on the ground. I have some ideas for enterprise but we'll see what happens there.

    I accidentally deleted your post editing my response on my phone too. :(

  • Thanks Janos, I'll bear that in mind for next time.
  • Thanks Guto, I'll give that a go. I commited a couple of fixes for the geotagging a while ago. It is getting better.

This reply was deleted.