Soil vs dry grass with multispectral images
Hello all,
I am attempting a 90% accuracy LC map with multispectral imagery and lidar. I usually don't have such difficulty with classification. The images were taken in the early spring and NIR other NIR derivatives (NDVI..) are not helpful in finding subtle differences in soil vs dry grass. I have shown an image capture. Pavement and dry soil are equally an issue. I find geometry and texture to be the most useful.The segments are small to keep classes closer to 90% accuracy..
Any one find a similar issue with a great solution?
Was this article helpful?
3 comments
Have you tried NDSI? I have found some success with this for delineating grass (lower values) and bare soil (higher values). The threshold depends on the imagery and the delineation between classes is not as obvious as NDVI, but could be worth exploring. I have found in general, thresholds of 0.075 or 0.1 (or so) worked well. Grass lower than the threshold. Soil above the threshold.
Hope this helps some.
Hello Cale Kockenour,
Thank you for the invaluable advice, it worked like a charm!
Very grateful
Glad to hear that it helped.