SAGA-GIS Tool Library Documentation (v6.2.0)

Tool [deprecated] Neural Networks (OpenCV)

Integration of the OpenCV Neural Networks library.
http://docs.opencv.org/modules/ml/doc/neural_networks.html

Parameters

 NameTypeIdentifierDescriptionConstraints
InputTrain INPUTTable (input)TRAIN_INPUT_TABLEInput data to train the network.-
Prediction INPUTTable (input)EVAL_INPUT_TABLEInput data for the trained network.-
Train INPUTGrid list (input)TRAIN_INPUT_GRIDSInput grids to train the network.-
Select training areasShapes (input)TRAIN_INPUT_AREASThe areas used to train the neural network.-
Sample weights (*)Table (optional input)WEIGHTSOptional floating-point vector of weights for each sample. Some samples may be more important than others for training.-
Indices (*)Table (optional input)INDICESOptional integer vector indicating the samples (rows of inputs and outputs) that are taken into account.-
OutputOUTPUTTable (output)EVAL_OUTPUT_TABLEPredicted responses (probabilities) for corresponding samples.-
OUTPUT classesGrid (output)EVAL_OUTPUT_GRID_CLASSESThese are the winner class vectors.-
OUTPUT certainty (*)Grid (optional output)EVAL_OUTPUT_GRID_CERTAINTYThese are the probabilities for the winner classes given by the neural network.-
OptionsGrid systemGrid systemPARAMETERS_GRID_SYSTEM--
Select training output featuresParametersTRAIN_FEATURES_TABLESelect the features that should be used as the OUTPUT features.0 Parameters:
Select class identifierTable fieldTRAIN_INPUT_AREAS_CLASS_FIELDThe class field used to classify the shape. All classes will be vectorized.-
Data typeChoiceDATA_TYPEChoose the type of data used in this model.Available Choices:
[0] Table
[1] Grid
Default: 0
Number of layersIntegerNNET_LAYERYou can specify the number of layers in the network (including input and output layer).Minimum: 1
Default: 3
Number of neuronsIntegerNNET_NEURONSYou can specify the number of neurons in each layer of the network.Minimum: 1
Default: 3
Maximum number of iterationsIntegerMAX_ITERTermination criteria of the training algorithm. You can specify the maximum number of iterations.Default: 1000
Error change (epsilon)Floating pointEPSILONTermination criteria of the training algorithm. You can specify how much the error could change between the iterations to make the algorithm continue (epsilon).Default: 0.010000
Activation functionChoiceACTIVATION_FUNCTIONChoose the activation function used by the neronal network.Available Choices:
[0] Identity
[1] Sigmoid
[2] Gaussian
Default: 1
AlphaFloating pointALPHAThis is the alpha parameter value for the sigmoid function.Default: 1.000000
BetaFloating pointBETAThis is the beta parameter value for the sigmoid function.Default: 1.000000
Training methodChoiceTRAINING_METHODChoose the training method used for the neronal network.Available Choices:
[0] Resilent propagation (RPROP)
[1] Backpropagation algorithm
Default: 1
Initial update valueFloating pointRP_DW0Initial value delta_0 of update-values delta_ij.Default: 0.000000
Increase factorFloating pointRP_DW_PLUSIncrease factor n_plus. Must be > 1.Minimum: 1.010000
Default: 1.200000
Decrease factorFloating pointRP_DW_MINUSDecrease factor n_minus. Must be < 1.Minimum: 0.010000
Maximum: 0.990000
Default: 0.500000
Update values lower limitFloating pointRP_DW_MINCorresponds to delta_min. Must be > 0Minimum: 0.010000
Default: 0.100000
Update values upper limitFloating pointRP_DW_MAXCorresponds to delta_max. Must be > 1Minimum: 1.010000
Default: 1.100000
Weight gradient termFloating pointBP_DW_SCALEStrength of the weight gradient term. The recommended value is about 0.1.Default: 0.100000
Moment termFloating pointBP_MOMENT_SCALEStrength of the momentum term (the difference between weights on the 2 previous iterations). This parameter provides some inertia to smooth the random fluctuations of the weights. It can vary from 0 (the feature is disabled) to 1 and beyond. The value 0.1 or so is good enough.Default: 0.100000
Update the network weightsBooleanUPDATE_WEIGHTS-Default: 0
No INPUT vector normalizationBooleanNO_INPUT_SCALEIf false, the training algorithm normalizes each input feature independently, shifting its mean value to 0 and making the standard deviation equal to 1.Default: 1
No OUTPUT vector normalizationBooleanNO_OUTPUT_SCALEIf false, the training algorithm normalizes each output feature independently, by transforming it to the certain range depending on the used activation function.Default: 1
(*) optional

Command-line

Usage: saga_cmd imagery_opencv 3 [-TRAIN_INPUT_TABLE <str>] [-EVAL_INPUT_TABLE <str>] [-EVAL_OUTPUT_TABLE <str>] [-TRAIN_INPUT_GRIDS <str>] [-TRAIN_INPUT_AREAS <str>] [-TRAIN_INPUT_AREAS_CLASS_FIELD <str>] [-EVAL_OUTPUT_GRID_CLASSES <str>] [-EVAL_OUTPUT_GRID_CERTAINTY <str>] [-DATA_TYPE <str>] [-NNET_LAYER <num>] [-NNET_NEURONS <num>] [-MAX_ITER <num>] [-EPSILON <double>] [-ACTIVATION_FUNCTION <str>] [-ALPHA <double>] [-BETA <double>] [-TRAINING_METHOD <str>] [-RP_DW0 <double>] [-RP_DW_PLUS <double>] [-RP_DW_MINUS <double>] [-RP_DW_MIN <double>] [-RP_DW_MAX <double>] [-BP_DW_SCALE <double>] [-BP_MOMENT_SCALE <double>] [-WEIGHTS <str>] [-INDICES <str>] [-UPDATE_WEIGHTS <str>] [-NO_INPUT_SCALE <str>] [-NO_OUTPUT_SCALE <str>]
  -TRAIN_INPUT_TABLE:<str>            	Train INPUT
	Table (input)
  -EVAL_INPUT_TABLE:<str>             	Prediction INPUT
	Table (input)
  -EVAL_OUTPUT_TABLE:<str>            	OUTPUT
	Table (output)
  -TRAIN_INPUT_GRIDS:<str>            	Train INPUT
	Grid list (input)
  -TRAIN_INPUT_AREAS:<str>            	Select training areas
	Shapes (input)
  -TRAIN_INPUT_AREAS_CLASS_FIELD:<str>	Select class identifier
	Table field
  -EVAL_OUTPUT_GRID_CLASSES:<str>     	OUTPUT classes
	Grid (output)
  -EVAL_OUTPUT_GRID_CERTAINTY:<str>   	OUTPUT certainty
	Grid (optional output)
  -DATA_TYPE:<str>                    	Data type
	Choice
	Available Choices:
	[0] Table
	[1] Grid
	Default: 0
  -NNET_LAYER:<num>                   	Number of layers
	Integer
	Minimum: 1
	Default: 3
  -NNET_NEURONS:<num>                 	Number of neurons
	Integer
	Minimum: 1
	Default: 3
  -MAX_ITER:<num>                     	Maximum number of iterations
	Integer
	Default: 1000
  -EPSILON:<double>                   	Error change (epsilon)
	Floating point
	Default: 0.010000
  -ACTIVATION_FUNCTION:<str>          	Activation function
	Choice
	Available Choices:
	[0] Identity
	[1] Sigmoid
	[2] Gaussian
	Default: 1
  -ALPHA:<double>                     	Alpha
	Floating point
	Default: 1.000000
  -BETA:<double>                      	Beta
	Floating point
	Default: 1.000000
  -TRAINING_METHOD:<str>              	Training method
	Choice
	Available Choices:
	[0] Resilent propagation (RPROP)
	[1] Backpropagation algorithm
	Default: 1
  -RP_DW0:<double>                    	Initial update value
	Floating point
	Default: 0.000000
  -RP_DW_PLUS:<double>                	Increase factor
	Floating point
	Minimum: 1.010000
	Default: 1.200000
  -RP_DW_MINUS:<double>               	Decrease factor
	Floating point
	Minimum: 0.010000
	Maximum: 0.990000
	Default: 0.500000
  -RP_DW_MIN:<double>                 	Update values lower limit
	Floating point
	Minimum: 0.010000
	Default: 0.100000
  -RP_DW_MAX:<double>                 	Update values upper limit
	Floating point
	Minimum: 1.010000
	Default: 1.100000
  -BP_DW_SCALE:<double>               	Weight gradient term
	Floating point
	Default: 0.100000
  -BP_MOMENT_SCALE:<double>           	Moment term
	Floating point
	Default: 0.100000
  -WEIGHTS:<str>                      	Sample weights
	Table (optional input)
  -INDICES:<str>                      	Indices
	Table (optional input)
  -UPDATE_WEIGHTS:<str>               	Update the network weights
	Boolean
	Default: 0
  -NO_INPUT_SCALE:<str>               	No INPUT vector normalization
	Boolean
	Default: 1
  -NO_OUTPUT_SCALE:<str>              	No OUTPUT vector normalization
	Boolean
	Default: 1