New answers tagged classification
One place to look for imagery sources is OpenStreetMap's "editor imagery index", which is basically a machine readable list of imagery provider URLs and extents. https://github.com/osmlab/editor-imagery-index The needs of your project aren't exactly the same as OSM's but there is overlap. In this case, in the Nairobi region it shows Bing and Mapbox ...
Not exactly "high-resolution," but the price is right. Esri's World Imagery map service presents satellite imagery for the world and high-resolution imagery for the United States and other areas around the world.
Majority Filter. You can apply multiple runs to smooth the image.
What I ended up doing: 1) Download and decompress your landsat tile. Do not open and look at the images in ArcMap because ArcMap will create .proj on the fly and this will confuse poor fmask and prevent it from running. 2) Once your landsat tile of choice's file is decompressed/unzipped/etc. Follow the directions on fmask's website for masking clouds. I ...
I found this at the QGIS 2.2 documentation at "Limitation for multi-band layers" Obviously there is a limitation of multi band layers, what means that they are not supported. As a work around one can extract every single band with the raster calculator Raster > Raster Calculator. Save each raster-band as a single tif-file an load it in the ...
I was looking for a possibility to delete isolated lines from OSM data, too. Unfortunately, I have a huge amount of data. And having to deal with Network Datasets anyway, I just didnt want to examine Geometric Networks. But using Michaels suggestion worked, thank you very much! I had to change 2 things: First, I had to unsplit the lines in the beginning, so ...
For anyone reading this, I eventually worked out a solution in PostGIS to select the nearest k neighbors (i.e. 9) and select the most common occurring class from the nearest neighbors. Then, I assigned that class to the white polygons. This approach was less sensitive to single, misplaced polygon classes.
After some research I came across a very promissing statistical approach: the Rand index. In R there are various implementations (package mclust & flexclust) - I will work them through in the next days and hopefully can deliver more information on the topic in a while.
Try setting "gruppo" = 'campione' in General | Query. Then do the classification. Then remove the query again. Hacky, but might work.
Random Forests in unlabeled (unsupervised) mode does not return explicit classes but, rather something analogous to scaled multivariate distances which is based on node proximities. Without the proximity matrix, you do not have a usable unlabeled model. And yes, for large problems, even using a sparse matrix, the very nature of the approach causes the ...
Top 50 recent answers are included