def train_stratum(self, stratum):
features_shapefile = self.get_output_path("features-{}.shp",
stratum.id)
split_features(stratum, self.args.refp, self.args.outdir)
area_training_polygons = self.get_output_path(
"training_polygons-{}.shp", stratum.id)
area_validation_polygons = self.get_output_path(
"validation_polygons-{}.shp", stratum.id)
area_statistics = self.get_output_path("statistics-{}.xml", stratum.id)
area_model = self.get_output_path("model-{}.txt", stratum.id)
area_confmatout = self.get_output_path(
"confusion-matrix-training-{}.csv", stratum.id)
area_days = self.get_output_path("days-{}.txt", stratum.id)
area_descriptors = []
area_prodpertile = []
for tile in stratum.tiles:
area_descriptors += tile.get_descriptor_paths()
area_prodpertile.append(len(tile.descriptors))
run_step(
Step("SampleSelection", [
"otbcli", "SampleSelection", self.args.buildfolder, "-ref",
features_shapefile, "-ratio", self.args.ratio, "-seed",
self.args.rseed, "-tp", area_training_polygons, "-vp",
area_validation_polygons
]))
step_args = [
"otbcli", "CropTypeTrainImagesClassifier", self.args.buildfolder,
"-mission", self.args.mission.name, "-nodatalabel", -10000,
"-pixsize", self.args.pixsize, "-outdays", area_days, "-mode",
self.args.trm, "-io.vd", area_training_polygons, "-rand",
self.args.rseed, "-sample.bm", 0, "-io.confmatout",
area_confmatout, "-io.out", area_model, "-sample.mt",
self.args.nbtrsample, "-sample.mv", 10, "-sample.vfn", "CODE",
"-sample.vtr", 0.01, "-classifier", self.args.classifier
]
if self.args.red_edge:
step_args += ["-rededge", "true"]
step_args += ["-sp"] + self.args.sp
step_args += ["-prodpertile"] + area_prodpertile
step_args += ["-il"] + area_descriptors
if self.args.classifier == "rf":
step_args += [
"-classifier.rf.nbtrees", self.args.rfnbtrees,
"-classifier.rf.min", self.args.rfmin, "-classifier.rf.max",
self.args.rfmax
]
else:
step_args += [
"-classifier.svm.k", "rbf", "-classifier.svm.opt", 1,
"-imstat", area_statistics
]
run_step(Step("TrainImagesClassifier", step_args))
def train_stratum(self, stratum):
area_model = self.get_output_path("model-{}.txt", stratum.id)
area_confmatout = self.get_output_path(
"confusion-matrix-training-{}.csv", stratum.id)
if self.args.refp is not None:
features_shapefile = self.get_output_path("features-{}.shp",
stratum.id)
split_features(stratum, self.args.refp, self.args.outdir)
area_training_polygons = self.get_output_path(
"training_polygons-{}.shp", stratum.id)
area_validation_polygons = self.get_output_path(
"validation_polygons-{}.shp", stratum.id)
area_statistics = self.get_output_path("statistics-{}.xml",
stratum.id)
area_days = self.get_output_path("days-{}.txt", stratum.id)
area_descriptors = []
area_prodpertile = []
for tile in stratum.tiles:
area_descriptors += tile.get_descriptor_paths()
area_prodpertile.append(len(tile.descriptors))
run_step(
Step("SampleSelection", [
"otbcli", "SampleSelection", self.args.buildfolder, "-ref",
features_shapefile, "-ratio", self.args.ratio, "-seed",
self.args.rseed, "-nofilter", "true", "-tp",
area_training_polygons, "-vp", area_validation_polygons
]))
step_args = [
"otbcli", "CropMaskTrainImagesClassifier",
self.args.buildfolder, "-progress", "false", "-mission",
self.args.mission.name, "-nodatalabel", -10000, "-pixsize",
self.args.pixsize, "-outdays", area_days, "-mode",
self.args.trm, "-io.vd", area_training_polygons, "-rand",
self.args.rseed, "-sample.bm", 0, "-io.confmatout",
area_confmatout, "-io.out", area_model, "-sample.mt",
self.args.nbtrsample, "-sample.mv", 10, "-sample.vfn", "CROP",
"-sample.vtr", 0.01, "-window", self.args.window, "-bm",
"true" if self.args.bm else "false", "-classifier",
self.args.classifier
]
if self.args.red_edge:
step_args += ["-rededge", "true"]
step_args += ["-sp"] + self.args.sp
step_args += ["-prodpertile"] + area_prodpertile
step_args += ["-il"] + area_descriptors
if self.args.classifier == "rf":
step_args += [
"-classifier.rf.nbtrees", self.args.rfnbtrees,
"-classifier.rf.min", self.args.rfmin,
"-classifier.rf.max", self.args.rfmax
]
else:
step_args += [
"-classifier.svm.k", "rbf", "-classifier.svm.opt", 1,
"-imstat", area_statistics
]
run_step(Step("TrainImagesClassifier", step_args))
else:
for tile in stratum.tiles:
tile_reference_trimmed = self.get_output_path(
"reference-trimmed-{}.tif", tile.id)
tile_stratum_reference_trimmed = self.get_output_path(
"reference-trimmed-{}-{}.tif", stratum.id, tile.id)
self.rasterize_tile_mask(stratum, tile)
stratum_tile_mask = self.get_stratum_tile_mask(stratum, tile)
step_args = [
"otbcli_BandMath", "-progress", "false", "-exp",
"im1b1 > 0 ? im2b1 : -10000", "-il", stratum_tile_mask,
tile_reference_trimmed, "-out",
format_otb_filename(tile_stratum_reference_trimmed,
compression='DEFLATE'), "int16"
]
run_step(Step("BandMath_" + str(tile.id), step_args))
area_model = self.get_output_path("model-{}.txt", stratum.id)
area_confmatout = self.get_output_path(
"confusion-matrix-training-{}.csv", stratum.id)
if self.args.classifier == "svm":
files = []
for tile in stratum.tiles:
tile_spectral_features = self.get_output_path(
"spectral-features-{}.tif", tile.id)
tile_stratum_spectral_features = self.get_output_path(
"spectral-features-{}-{}.tif", stratum.id, tile.id)
stratum_tile_mask = self.get_stratum_tile_mask(
stratum, tile)
step_args = [
"otbcli_BandMath", "-progress", "false", "-exp",
"im1b1 > 0 ? im2b1 : -10000", "-il", stratum_tile_mask,
tile_spectral_features, "-out",
format_otb_filename(tile_stratum_spectral_features,
compression='DEFLATE'), "int16"
]
run_step(Step("BandMath_" + str(tile.id), step_args))
files.append(tile_stratum_spectral_features)
step_args = [
"otbcli_ComputeImagesStatistics", "-bv", -10000, "-out",
area_statistics, "-il"
] + files
step_args = [
"otbcli", "TrainImagesClassifierNew", self.args.buildfolder,
"-nodatalabel", -10000, "-rand", self.args.rseed, "-sample.bm",
0, "-io.confmatout", area_confmatout, "-io.out", area_model,
"-sample.mt", self.args.nbtrsample, "-sample.mv", 1000,
"-sample.vfn", "CROP", "-sample.vtr", 0.01, "-classifier",
self.args.classifier
]
if self.args.classifier == "rf":
step_args += [
"-classifier.rf.nbtrees", self.args.rfnbtrees,
"-classifier.rf.min", self.args.rfmin,
"-classifier.rf.max", self.args.rfmax
]
else:
step_args += [
"-classifier.svm.k", "rbf", "-classifier.svm.opt", 1,
"-imstat", area_statistics
]
step_args.append("-io.rs")
for tile in stratum.tiles:
tile_stratum_reference_trimmed = self.get_output_path(
"reference-trimmed-{}-{}.tif", stratum.id, tile.id)
step_args.append(tile_stratum_reference_trimmed)
step_args.append("-io.il")
for tile in stratum.tiles:
tile_spectral_features = self.get_output_path(
"spectral-features-{}.tif", tile.id)
step_args.append(tile_spectral_features)
run_step(Step("TrainImagesClassifier", step_args))