I have 10 2D geoTIFF orthophotos (~2GB each, ~3km2 each, all 4.2cm/pixel) that I generated from Vis RGB 12MP images in Pix4D for a desert vegetation survey over 30km2 for a development. I need to calculate the veg cover under a polygon in each ortho. After a bit of searching I opted for the SCP plugin by Luca Congedo. It seemed to work a treat, I can understand the UI just enough to get it to work and can overlay my mask layer to only classify that part.

Trouble is for one polygon is has taken often days to process, often fails, and the resulting classified shapes have over 1M features, making area calculation laggy and frankly I haven't managed to get very far with it. I've even divided each mask into 4 to reduce the area to classify and it takes almost the same amount of time for a quarter of the area.

Eg. I completed one small area (probably 1/50 of the total area), took 22 hours, resulting .shp is 443MB with 1.5M features and is practically unworkable due to lag for any type of action. I removed the non-veg classes I'm not interested in and Save As'd and it has so far taken 1 hour and counting...

I'm using 2 identical Dell XPS 9550's: quad-core i7 6700HQ 2.6 GHz; 2133MHz 32GB RAM; 512GB SSD; NVIDIA GTX960M 2GB; so no room for change there. Although interestingly neither my CPU, RAM or GPU ever get close to maxing out, so I'm thinking the plugin just isn't optimized for large high res images.

Can anyone suggest a faster method using QGIS? Or maybe there's some setting in SCP that I need to tweak?

screenshot of veg type1

scrfeenshot of veg type2

  • Could you please post a screenshot highlighting the type of vegetation you are looking to classify? Are you looking to classify any green vegetation and leave everything else? Any chance you work with Python or R?
    – Aaron
    Jan 1 '18 at 5:44
  • all the veg is grey, never worked with Python or R, i'm a newbie, 25 years since i last touched GIS Jan 1 '18 at 5:58
  • essentially all i need to do is pick out the grey veg patches and determine their combined area. the colors don't help, there are a lot of other artifacts with similar tone, but once classified eliminating the poligons above and below a certain size eliminated a lot of the noise, and a coarse estimate is all i need (from 30km2!) Jan 1 '18 at 6:12
  • To begin, I would keep this a raster problem rather than converting to vector--it is much more efficient this way. Try integrating 3 band veg indices into your classification (more details here: digitalcommons.unl.edu/cgi/…). I've also had success using texture measures for problems such as yours.
    – Aaron
    Jan 1 '18 at 6:27

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.