i2_classification

Landsat5_old

Name	Default Value	Description	Type	Mandatory	Name
additional_features		OTB’s bandmath expressions, separated by comma	str	False	additional_features
end_date		The end date of interpolated image time series	str	False	end_date
keep_bands	[‘B1’, ‘B2’, ‘B3’, ‘B4’, ‘B5’, ‘B6’]	The list of spectral bands used for classification	list	False	keep_bands
start_date		The first date of interpolated image time series	str	False	start_date
temporal_resolution	16	The temporal gap between two interpolation	int	False	temporal_resolution

Landsat8

Name	Default Value	Description	Type	Mandatory	Name
additional_features		OTB’s bandmath expressions, separated by comma	str	False	additional_features
end_date		The end date of interpolated image time series	str	False	end_date
keep_bands	[‘B1’, ‘B2’, ‘B3’, ‘B4’, ‘B5’, ‘B6’, ‘B7’]	The list of spectral bands used for classification	list	False	keep_bands
start_date		The first date of interpolated image time series	str	False	start_date
temporal_resolution	16	The temporal gap between two interpolation	int	False	temporal_resolution
write_reproject_resampled_input_dates_stack	True	Enable the write of resampled stack image for each date	bool	False	write_reproject_resampled_input_dates_stack

Landsat8_old

Name	Default Value	Description	Type	Mandatory	Name
additional_features		OTB’s bandmath expressions, separated by comma	str	False	additional_features
end_date		The end date of interpolated image time series	str	False	end_date
keep_bands	[‘B1’, ‘B2’, ‘B3’, ‘B4’, ‘B5’, ‘B6’]	The list of spectral bands used for classification	list	False	keep_bands
start_date		The first date of interpolated image time series	str	False	start_date
temporal_resolution	16	The temporal gap between two interpolation	int	False	temporal_resolution
write_reproject_resampled_input_dates_stack	True	Enable the write of resampled stack image for each date	bool	False	write_reproject_resampled_input_dates_stack

Sentinel_2

Name	Default Value	Description	Type	Mandatory	Name
additional_features		OTB’s bandmath expressions, separated by comma	str	False	additional_features
end_date		The end date of interpolated image time series	str	False	end_date
keep_bands	[‘B2’, ‘B3’, ‘B4’, ‘B5’, ‘B6’, ‘B7’, ‘B8’, ‘B8A’, ‘B11’, ‘B12’]	The list of spectral bands used for classification	list	False	keep_bands
start_date		The first date of interpolated image time series	str	False	start_date
temporal_resolution	10	The temporal gap between two interpolation	int	False	temporal_resolution
write_reproject_resampled_input_dates_stack	True	Enable the write of resampled stack image for each date	bool	False	write_reproject_resampled_input_dates_stack

Sentinel_2_L3A

Name	Default Value	Description	Type	Mandatory	Name
additional_features		OTB’s bandmath expressions, separated by comma	str	False	additional_features
end_date		The end date of interpolated image time series	str	False	end_date
keep_bands	[‘B2’, ‘B3’, ‘B4’, ‘B5’, ‘B6’, ‘B7’, ‘B8’, ‘B8A’, ‘B11’, ‘B12’]	The list of spectral bands used for classification	list	False	keep_bands
start_date		The first date of interpolated image time series	str	False	start_date
temporal_resolution	16	The temporal gap between two interpolation	int	False	temporal_resolution
write_reproject_resampled_input_dates_stack	True	Enable the write of resampled stack image for each date	bool	False	write_reproject_resampled_input_dates_stack

Sentinel_2_S2C

Name	Default Value	Description	Type	Mandatory	Name
additional_features		OTB’s bandmath expressions, separated by comma	str	False	additional_features
end_date		The end date of interpolated image time series	str	False	end_date
keep_bands	[‘B02’, ‘B03’, ‘B04’, ‘B05’, ‘B06’, ‘B07’, ‘B08’, ‘B8A’, ‘B11’, ‘B12’]	The list of spectral bands used for classification	list	False	keep_bands
start_date		The first date of interpolated image time series	str	False	start_date
temporal_resolution	10	The temporal gap between two interpolation	int	False	temporal_resolution
write_reproject_resampled_input_dates_stack	True	Enable the write of resampled stack image for each date	bool	False	write_reproject_resampled_input_dates_stack

arg_classification

Name	Default Value	Description	Type	Mandatory	Name
classif_mode	separate	‘separate’ or ‘fusion’.	str	False	classif_mode
dempstershafer_mob	precision	Choose the dempster shafer mass of belief estimation method	str	False	dempstershafer_mob
enable_probability_map	False	Produce the probability map	bool	False	enable_probability_map
fusion_options	-nodatalabel 0 -method majorityvoting	OTB FusionOfClassification options for voting method involved if classif_mode is set to ‘fusion’	str	False	fusion_options
fusionofclassification_all_samples_validation	False	Enable the use of all reference data	bool	False	fusionofclassification_all_samples_validation
keep_runs_results	True		bool	False	keep_runs_results
merge_final_classifications	False	Enable the fusion of classifications mode, merging all run in a unique result.	bool	False	merge_final_classifications
merge_final_classifications_method	majorityvoting	Indicate the fusion of classification method: ‘majorityvoting’ or ‘dempstershafer’	str	False	merge_final_classifications_method
merge_final_classifications_ratio	0.1	Percentage of samples to use in order to evaluate the fusion raster	float	False	merge_final_classifications_ratio
merge_final_classifications_undecidedlabel	255	Indicate the label for undecision case during fusion	int	False	merge_final_classifications_undecidedlabel
no_label_management	maxConfidence	Method for choosing a label in case of fusion	str	False	no_label_management

Notes

classif_mode

If ‘fusion’ : too huge models will be devided into smaller ones and will classify the same pixels. The treshold between small/big models is define by the parameter ‘mode_outside_regionsplit’

dempstershafer_mob

Two kind of indexes can be used: * Global: accuracy or kappa * Per class: precision or recall

enable_probability_map

A probability map is a image with N bands , where N is the number of classes in the nomenclature file. The bands are sorted in ascending order more information more information here

fusionofclassification_all_samples_validation

If the fusion mode is enabled, enable the use of all reference data samples for validation

keep_runs_results

If in fusion mode, two final reports can be provided. One for each seed, and one for the classification fusion

arg_train

Name	Default Value	Description	Type	Mandatory	Name
a_crop_label_replacement	[‘10’, ‘annual_crop’]	Replace a label by a string	list	False	a_crop_label_replacement
annual_classes_extraction_source	None		str	False	annual_classes_extraction_source
annual_crop	[‘11’, ‘12’]	The list of classes to be replaced by previous data	list	False	annual_crop
autocontext_iterations	3	Number of iterations in auto-context.	int	False	autocontext_iterations
classifier	None	Choose the classification algorithm	str	False	classifier
crop_mix	False	Enable crop mix option	bool	False	crop_mix
deep_learning_parameters	{}	deep learning parameter description is available here	dict	False	deep_learning_parameters
dempster_shafer_sar_opt_fusion	False	Enable the use of both SAR and optical data to train a model.	bool	False	dempster_shafer_sar_opt_fusion
enable_autocontext	False	Enable the auto-context processing	bool	False	enable_autocontext
features	[‘NDVI’, ‘NDWI’, ‘Brightness’]	List of additional features computed	list	False	features
features_from_raw_dates	False	learn model from raw sensor’s date (no interpolations)	bool	False	features_from_raw_dates
force_standard_labels	False	Standardize labels for feature extraction	bool	False	force_standard_labels
mode_outside_regionsplit	0.1	Fix the threshold for split huge model	float	False	mode_outside_regionsplit
otb_classifier_options	None	OTB option for classifier. If None, the OTB default values are used.	dict	False	otb_classifier_options
output_prev_features	None	Path to previous features for crop mix	str	False	output_prev_features
prev_features	None	Path to a configuration file used to produce previous features	str	False	prev_features
random_seed	None	Fix the random seed for random split of reference data	int	False	random_seed
ratio	0.5	Should be between 0.0 and 1.0 and represent the proportion of the dataset to include in the train split.	float	False	ratio
runs	1	Number of independant runs processed.	int	False	runs
sample_augmentation	{‘activate’: False}	OTB parameters for sample augmentation	dict	False	sample_augmentation
sample_management	None	Absolute path to a CSV file containing samples transfert strategies	str	False	sample_management
sample_selection	{‘sampler’: ‘random’, ‘strategy’: ‘all’}	OTB parameters for sample selection	dict	False	sample_selection
sample_validation	{‘sampler’: ‘random’, ‘strategy’: ‘all’}	OTB parameters for sampling the validation set	dict	False	sample_validation
samples_classif_mix	False	Enable the second step of crop mix	bool	False	samples_classif_mix
sampling_validation	False	Enable sampling validation	bool	False	sampling_validation
split_ground_truth	True	Enable the split of reference data	bool	False	split_ground_truth
validity_threshold	1		int	False	validity_threshold

Notes

dempster_shafer_sar_opt_fusion

Enable the use of both SAR and optical data to train a model. If True then two models are trained.more documentation is avalailalbe here

features

This parameter enable the computation of the three indices if available for the sensor used.There is no choice for using only one of them.

features_from_raw_dates

If True, during the learning and classification step, each pixel will receive a vector of values of the size of the number of all dates detected. As the pixels were not all acquired on the same dates, the vector will contains NaNs on the unacquired dates. In iota2, we have chosen to keep these NaNs. The responsibility of managing these NaNs is delegated to the classification system (e.g. the forward method in deepLearning). Furthermore, in this use case, binary masks are provided to the classification system. These masks contains 3 types of information for 3 differents pixel values

• 0 : pixel acquired by the sensor and valid

• 1 : pixel acquired by the sensor and not valid (cloud,…)

• 2 : pixel not acquired.

force_standard_labels

The chain label each features by the sensors name, the spectral band or indice and the date. If activated this parameter use the OTB default value (value_X)

mode_outside_regionsplit

This parameter is available if regionPath is used and arg_classification.classif_mode is set to fusion. It represents the maximum size covered by a region. If the regions are larger than this threshold, then N models are built by randomly selecting features inside the region.

otb_classifier_options

This parameter is a dictionnary which accepts all OTB application parameters. To know the exhaustive parameter list use otbcli_TrainVectorClassifier in a terminal or look at the OTB documentation

prev_features

This config file must be launchable by iota2 (needed for crop mix)

random_seed

Fix the random seed used for random split of reference data If set, the results must be the same for a given classifier

runs

Number of independant runs processed. Each run has his own learning samples. Must be an integer greater than 0

sample_augmentation

In supervised classification the balance between class samples is important. There are any ways to manage class balancing in iota2, using sample_selection or the classifier’s options to limit the number of samples by class. An other approch is to generate synthetic samples. It is the purpose of thisfunctionality, which is called ‘sample augmentation’.

{'activate':False}

Example

sample_augmentation : {'target_models':['1', '2'],
                      'strategy' : 'jitter',
                      'strategy.jitter.stdfactor' : 10,
                      'strategy.smote.neighbors'  : 5,
                      'samples.strategy' : 'balance',
                      'activate' : True
                      }

iota2 implements an interface to the OTB SampleAugmentation application. There are three methods to generate samples : replicate, jitter and smote.The documentation here explains the difference between these approaches.

samples.strategy specifies how many samples must be created.There are 3 different strategies:

• minNumber

  To set the minimum number of samples by class required

• balance

  balance all classes with the same number of samples as the majority one

• byClass

  augment only some of the classes

Parameters related to minNumber and byClass strategies are:

• samples.strategy.minNumber

  minimum number of samples

• samples.strategy.byClass

  path to a CSV file containing in first column the class’s label and in the second column the minimum number of samples required.

In the above example, classes of models ‘1’ and ‘2’ will be augmented to thethe most represented class in the corresponding model using the jitter method.

sample_management

The CSV must contain a row per transfert

>>> cat /absolute/path/myRules.csv
    1,2,4,2

Meaning :

source	destination	class name	quantity
1	2	4	2

sample_selection

This field parameters the strategy of polygon sampling. It directly refers to options of OTB’s SampleSelection application.

Example

sample_selection : {'sampler':'random',
                   'strategy':'percent',
                   'strategy.percent.p':0.2,
                   'per_models':[{'target_model':'4',
                                  'sampler':'periodic'}]
                   }

In the example above, all polygons will be sampled with the 20% ratio. But the polygons which belong to the model 4 will be periodically sampled, instead of the ransom sampling used for other polygons. Notice than per_models key contains a list of strategies. Then we can imagine the following :

sample_selection : {'sampler':'random',
                   'strategy':'percent',
                   'strategy.percent.p':0.2,
                   'per_models':[{'target_model':'4',
                                  'sampler':'periodic'},
                                 {'target_model':'1',
                                  'sampler':'random',
                                  'strategy', 'byclass',
                                  'strategy.byclass.in', '/path/to/myCSV.csv'
                                 }]
                   }

where the first column of /path/to/myCSV.csv is class label (integer), second one is the required samples number (integer).

sample_validation

This field parameters the strategy of polygon sampling. It directly refers to options of OTB’s SampleSelection application.

Example

sample_validation : {'sampler':'random',
                   'strategy':'all',
                   }

split_ground_truth

If set to False, the chain use all polygons for both training and validation

builders

Name	Default Value	Description	Type	Mandatory	Name
builders_class_name	[‘i2_classification’]	The name of the class defining the builder	list	False	builders_class_name
builders_paths	/path/to/iota2/sources	The path to user builders	list	False	builders_paths

Notes

builders_class_name

Available builders are : ‘i2_classification’, ‘i2_features_map’, ‘i2_obia’ and ‘i2_vectorization’

builders_paths

If not indicated, the iota2 source directory is used: */iota2/sequence_builders/

chain

Name	Default Value	Description	Type	Mandatory	Name
check_inputs	True	Enable the inputs verification.	bool	False	check_inputs
cloud_threshold	0	Threshold to consider that a pixel is valid	int	False	cloud_threshold
color_table	None	Absolute path to the file which link classes and their colors	str	True	color_table
data_field	None	Field name indicating classes labels in ground_thruth	str	True	data_field
first_step	None	The step group name indicating where the chain start	str	True	first_step
ground_truth	None	Absolute path to reference data	str	True	ground_truth
l5_path_old	None	Absolute path to Landsat-5 images coming from old THEIA format (D*H*)	str	False	l5_path_old
l8_path	None	Absolute path to Landsat-8 images comingfrom new tiled THEIA data	str	False	l8_path
l8_path_old	None	Absolute path to Landsat-8 images coming from old THEIA format (D*H*)	str	False	l8_path_old
last_step	None	The step group name indicating where the chain ends	str	True	last_step
list_tile	None	List of tile to process, separated by space	str	True	list_tile
logger_level	INFO	Set the logger level: NOTSET, DEBUG, INFO, WARNING, ERROR, CRITICAL	str	False	logger_level
minimum_required_dates	2	required minimum number of available dates for each sensor	int	False	minimum_required_dates
nomenclature_path	None	Absolute path to the nomenclature description file	str	True	nomenclature_path
output_path	None	Absolute path to the output directory.	str	True	output_path
output_statistics	False	Enable the writing of PNG files containing additional statistics	bool	False	output_statistics
proj	EPSG:2154	The projection wanted. Format EPSG:XXXX is mandatory	str	False	proj
region_field	region	The column name for region indicator in`region_path` file	str	False	region_field
region_path	None	Absolute path to region vector file	str	False	region_path
remove_output_path	True	Enable the removing of complete output_path directory	bool	False	remove_output_path
s1_path	None	Absolute path to Sentinel-1 configuration file	str	False	s1_path
s2_l3a_output_path	None	Absolute path to store preprocessed data in a dedicated directory.	str	False	s2_l3a_output_path
s2_l3a_path	None	Absolute path to Sentinel-2 L3A images (THEIA format)	str	False	s2_l3a_path
s2_output_path	None	Absolute path to store preprocessed data in a dedicated directory.	str	False	s2_output_path
s2_path	None	Absolute path to Sentinel-2 images (THEIA format)	str	False	s2_path
s2_s2c_output_path	None	Absolute path to store preprocessed data in a dedicated directory.	str	False	s2_s2c_output_path
s2_s2c_path	None	Absolute path to Sentinel-2 images (Sen2Cor format)	str	False	s2_s2c_path
spatial_resolution	10	Output spatial resolution	list	False	spatial_resolution
user_feat_path	None	Absolute path to the user’s features path	str	False	user_feat_path

Notes

check_inputs

Enable the inputs verification. It can take a lot of time for large dataset. Check if region intersect reference data for instance

cloud_threshold

Indicates the threshold for a polygon to be used for learning. It use the validity count, which is incremented if a cloud, a cloud shadow or a saturated pixel is detected

output_path

Absolute path to the output directory.It is recommended to have one directory per run of the chain

output_statistics

Enable the writing of PNG files containing additional statisticscontaining the confidence by learning/validation pixels

region_field

this column in the database must contains string which can be converted into integers. For instance ‘1_2’ does not match this condition

remove_output_path

Enable the removing of complete output_path directory Only if the first_step is init and the folder name is valid

spatial_resolution

The spatial resolution expected.It can be provided as integer or float,or as a list containing two values for non squared resolution

user_feat_path

Absolute path to the user’s features path They must be stored by tiles

coregistration

Name	Default Value	Description	Type	Mandatory	Name
band_ref	1	Number of the band of the VHR image to use for coregistration	int	False	band_ref
band_src	3	Number of the band of the src raster to use for coregistration	int	False	band_src
date_src	None	Date YYYYMMDD of the reference image	str	False	date_src
date_vhr	None	Date YYYYMMDD of the VHR image	str	False	date_vhr
iterate	True	Proceed several iteration by reducing the step between geobin to find SIFT points	bool	False	iterate
minsiftpoints	40	Minimal number of SIFT points to find to create the new RPC model	int	False	minsiftpoints
minstep	16	Minimal size of steps between bins in pixels	int	False	minstep
mode	2	Coregistration mode of the time series	int	False	mode
pattern	None	Pattern of the time series files to coregister	str	False	pattern
prec	3	Estimated shift between source and reference raster in pixel (source raster resolution)	int	False	prec
resample	True	Resample the reference and the source rasterto the same resolution to find SIFT points	bool	False	resample
step	256	Initial size of steps between bins in pixels	int	False	step
vhr_path	None	Absolute path to VHR path	str	False	vhr_path

Notes

date_src

If no date_src is mentionned, the best image will be automatically choose for coregistration

mode

Mode	Method
1	single coregistration between one source image (and its masks) and the VHR image
2	this mode operates a coregistration between a image of the timeseries and the VHR image, then the same RPC model is used to orthorectify every images of the timeseries
3	cascade mode, this mode operates a first coregistration between a source image and the VHR image, then each image of the timeseries is coregistered step by step with the closest temporal images of the timeseries already coregistered

pattern

By default the value is left to None and the pattern depends on the sensor used (STACK.tif for Sentinel2, ORTHO_SURF_CORR_PENTE.TIF)

Examples

pattern: '*STACK.tif'

dim_red

Name	Default Value	Description	Type	Mandatory	Name
dim_red	False	Enable the dimensionality reduction mode	bool	False	dim_red
reduction_mode	global	The reduction mode	str	False	reduction_mode
target_dimension	4	The number of dimension required, according to reduction_mode	int	False	target_dimension

Notes

reduction_mode

Values authorized are: ‘global’ or ‘?’

external_features

Name	Default Value	Description	Type	Mandatory	Name
concat_mode	True	Enable the use of all features	bool	False	concat_mode
exogeneous_data	None	Path to a Geotiff file containing additional data to be used in external features.	str	False	exogeneous_data
external_features_flag	False	Enable the external features mode	bool	False	external_features_flag
functions	None	The function list to be used to compute features.Can be a string of space-separated function namesCan be a list of either strings of function nameor lists of one function name and one argument mapping	list	False	functions
module	/path/to/iota2/sources	Absolute path for user source code	str	False	module
output_name	None	Temporary chunks are written using this name as prefix	str	False	output_name

Notes

concat_mode

If disabled, only external features are used in the whole processing

exogeneous_data

If the =exogeneous_data= contains ‘$TILE’, it will be replaced by the tile name being processed.If you want to reproject your data on given tiles, you can use the =split_raster_into_tiles.py= command line tool.

Usage: =split_raster_into_tiles.py –help=.

iota2_feature_extraction

Name	Default Value	Description	Type	Mandatory	Name
acor_feat	False	Apply atmospherically corrected features	bool	False	acor_feat
copy_input	True	use spectral bands as features	bool	False	copy_input
extract_bands	False		bool	False	extract_bands
keep_duplicates	True	use ‘rel_refl’ can generate duplicated feature (ie: NDVI), set to False remove these duplicated features	bool	False	keep_duplicates
rel_refl	False	Compute relative reflectances by the red band	bool	False	rel_refl

Notes

acor_feat

Apply atmospherically corrected featuresas explained at : http://www.cesbio.ups-tlse.fr/multitemp/?p=12746

python_data_managing

Name	Default Value	Description	Type	Mandatory	Name
chunk_size_mode	split_number	The chunk split mode, choices are ‘split_number’ or ‘user_fixed’	str	False	chunk_size_mode
chunk_size_x	50	The number if rows for chunk	int	False	chunk_size_x
chunk_size_y	50	The number if rows for chunk	int	False	chunk_size_y
data_mode_access	gapfilled	Choose which data can be accessed in custom features	str	False	data_mode_access
fill_missing_dates	False	Fill raw data with no data if dates are missing	bool	False	fill_missing_dates
max_nn_inference_size	None	maximum batch inference size	int	False	max_nn_inference_size
number_of_chunks	50	The expected number of chunks	int	False	number_of_chunks
padding_size_x	0	The padding for chunk	int	False	padding_size_x
padding_size_y	0	The padding for chunk	int	False	padding_size_y

Notes

data_mode_access

Three values are allowed: - gapfilled: give access only the gapfilled data - raw: gives access only the original raw data - both: provides access to both data ..Notes:: Data are spatialy resampled, these parameters concern only temporal interpolation

fill_missing_dates

If raw data access is enabled, this option considers all unique dates for all tiles and identify which dates are missing for each tile. A missing date is filled using a no data constant value.Cloud or saturation are not corrected, but masks are provided Masks contain three value: 0 for valid data, 1 for cloudy or saturated pixels, 2 for a missing date

max_nn_inference_size

Involved if a neural network inference is performed. If not set (None), the inference size will be the same as the one used during the learning stage

scikit_models_parameters

Name	Default Value	Description	Type	Mandatory	Name
cross_validation_folds	5	The number of k-folds	int	False	cross_validation_folds
cross_validation_grouped	False		bool	False	cross_validation_grouped
cross_validation_parameters	{}		dict	False	cross_validation_parameters
model_type	None	machine learning algorthm’s name	str	False	model_type
standardization	False		bool	False	standardization

Notes

model_type

Models comming from scikit-learn are use if scikit_models_parameters.model_type is different from None. More informations about how to use scikit-learn is available at iota2 and scikit-learn machine learning algorithms.

sensors_data_interpolation

Name	Default Value	Description	Type	Mandatory	Name
auto_date	True	Enable the use of start_date and end_date	bool	False	auto_date
use_additional_features	False	Enable the use of additional features	bool	False	use_additional_features
use_gapfilling	True	Enable the use of gapfilling	bool	False	use_gapfilling
write_outputs	False	Write temporary files	bool	False	write_outputs

Notes

auto_date

If True, iota2 will automatically guess the first and the last interpolation date. Else, start_date and end_date of each sensors will be used

write_outputs

Write the time series before and after gapfilling, the mask time series, and also the feature time series. This option required a large amount of free disk space.

task_retry_limits

Name	Default Value	Description	Type	Mandatory	Name
allowed_retry	0	Allow dask to retry a failed job N times.	int	False	allowed_retry
maximum_cpu	4	The maximum number of CPU available	int	False	maximum_cpu
maximum_ram	16	The maximum amount of RAM available. (gB)	int	False	maximum_ram

Notes

maximum_cpu

the amount of cpu will be doubled if the task is killed due to ram overconsumption until maximum_cpu or allowed_retry are reach

maximum_ram

the amount of RAM will be doubled if the task is killed due to ram overconsumption until maximum_ram or allowed_retry are reach

userFeat

Name	Default Value	Description	Type	Mandatory	Name
arbo	/*	The input folder hierarchy	str	False	arbo
patterns	ALT,ASP,SLP	key name for detect the input images	str	False	patterns