User Tools

Site Tools


wiki:exercise8a

Exercise 8a: Image classification: create training sample and masks

pktools: pkinfo, pkmosaic, pksetmask, pkextract

Create mosaic from FMAP2006 for same coverage:

pkmosaic $(pkinfo -cover $(for file in ${FMAPDIR}/CM-FMAP_2006_*-AA.tif;do echo " -i $file";done) $(pkinfo -bb -i ${LANDSATDIR}/${LANDSATIMG})) $(pkinfo -bb -i ${LANDSATDIR}/${LANDSATIMG} -dx -dy) -o ${OUTPUTDIR}/exercise8/fmap2006.tif

Merge all masks into single mask (start with cloud mask and add other masks). Recode mask values as follows:

notice that cloudmask is not only input but also mask!
pksetmask -i ${OUTPUTDIR}/exercise6/cloudmask_dil.tif -m ${OUTPUTDIR}/exercise6/cloudmask_dil.tif -msknodata 1 -nodata 255 -m ${OUTPUTDIR}/exercise8/fmap2006.tif -msknodata 251 -nodata 251 -m ${OUTPUTDIR}/exercise6/shadowmask.tif -msknodata 1 -nodata 254 -o ${OUTPUTDIR}/exercise8/mask.tif -ct ${FMAPDIR}/ct_ftyp.txt
sea: 251
shadow: 254
cloud: 255

Create a self-sufficient training sample for classifier: output vector contains both label and spectral information (can take a wile, approx 20 min)

pkextract -i ${LANDSATDIR}/${LANDSATIMG} -s ${OUTPUTDIR}/exercise7/osm_merged.sqlite -o ${OUTPUTDIR}/exercise8/training.sqlite -r mean -f SQLite -m ${OUTPUTDIR}/exercise8/mask.tif -msknodata 251 -msknodata 254 -msknodata 255

Classify Landsat image using training sample and masks: sea (251), shadow (254), clouds (255)

pkclassify supports two modes for training samples:
1. automatic mode: vector file has an Integer attribute representing the classes in a one-to-one relation (e.g., '1'= class 1, '2'= class 2)
2. manual mode: vector file has a String attribute representing the classes in a many-to-one relation (e.g., 'forest'= class 1, 'grass'= class 2, 'residential'= class 2)

Classification in manual mode:

pkclassify_svm -i ${LANDSATDIR}/${LANDSATIMG} -t ${OUTPUTDIR}/exercise8/training.sqlite -m ${OUTPUTDIR}/exercise8/mask.tif -msknodata 251 -msknodata 254 -msknodata 255 -o ${OUTPUTDIR}/exercise8/19990724_L7E_IM_FMAP.tif -label landuse -c commercial -r 2 -c forest -r 1 -c grass -r 2 -c industrial -r 2 -c meadow -r 2 -c residential -r 2 -ct ${OUTPUTDIR}/exercise8/ct_ftyp.txt
-msknodata values will be retained in classification output as “normal values”. You can set a real nodata value using option -nodata (default value = 0).
wiki/exercise8a.txt · Last modified: 2017/12/05 22:53 (external edit)