I notice on the web site page there is an image of field boundaries. Does anyone have a “recipe” for producing field boundaries using Orfeo with Sentinel-2 images?
Thanks!
Hi,
You should check the Segmentation recipe. Once you have vectorized the output regions, you can open the shapefile/sqlite in QGIS and display it over the initial Sentinel 2 image.
Thanks I had arrived there. But the results I was achieving were not great. I was using summed together seven or eight NIR Band images and then running the segmentation on that. I was hoping that someone might have a view on the input and any parameter changes to the defaults that might generate good results.
What do you mean “summed together”? making a stack ?
Generally for Segmentation, you want to tune the spatial and spectral ranges first. Then you use the minimum region size to filter small objects.
I found eight cloudless days I did gdal_calc.py -A input1.tif -B input2.tif --outfile=result.tif --calc=“A+B”
To build one B08 image with all the days added together.
I’m trying to understand how I should be going about this with Orfeo. I’ve read a lot of papers and basically the best results seem to come from combining temporarily several images. My hope was someone would have a set of instructions to do this. So what should the input be a colour Sentinel-2 image? Or what I am using or what? And then what parameters do you use for each of the tools, if not the defaults.
Thanks
The input of the segmentation algorithm is expected to be a multispectral image.
If you already have cloudless images, fusing them is maybe not usefull for segmentation. The multi-temporal information mostly helps to:
- fill areas with clouds
- analyse land cover ( vegetation types)