2

I want to classify 20 feet-resolution 4-band leaf-off image into lowland hardwood, upland hardwood, pine trees, and other (bare ground, water, roads etc). I am thinking to use randomForest package in R and classify the image in 3 classes (hardwood, pine trees, and other) and then use stream buffers to separate lowland and upland hardwood. Would it improve classification to add layers with NDVI, DEM or/and topographic wetness index? Which other layers should I try?

  • Could you add a bit more information as to what you have in terms of training data, what is the imagery source, etc. For what it is worth, adding more data is never a bad idea (within reason), and your models will tell you if something doesn't help. I would doubt that NDVI on it's own would help much with tree types. SAR polarimetry might, though. – John Powell Apr 21 '17 at 10:09
  • 1
    I like to incorporate NDVI and texture metrics along with the spectral bands when using 4-band imagery. I've also had success increasing classification accuracy by using a hierarchical classification approach, where I include binary classifications in the model. Incorporating landform data is always a good idea--here is a land cover classification competition using only landform data: kaggle.com/c/forest-cover-type-prediction. – Aaron Apr 21 '17 at 15:30
  • @JohnPowellakaBarça. Thank you, I'll try different options to see what works. I am going to create a shp-files with training data manually: the resolution is high enough so I can see where is pine trees, hardwood etc. Imagery source: aerial photo, sensor is Leica ADS80-SH82. – Lybica Apr 24 '17 at 15:50
  • @Aaron for your input. Binary classification seems like a good idea though i am not sure yet which metric would be useful. – Lybica Apr 24 '17 at 15:54

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.