Guidances in order to exploit temporal/seasonal ndvi/spectral variations with iota2 (nomenclatures with different levels)
Good afternoon, I'm taking the liberty to share you my classification context. May be you could express some advices/guidances ? My goal is to produce a land use mapping of the Dakar metropolitan (only 1 sentinel2 tile necessary). I have 205 samples. Regarding the climate : 1 wet season and 1 dry season. I share my classification iota2 current result with a dense time series (19 Sentinel2 images from January 2023 to today). The Nomenclature level 1 with 4 classes :
- Artificialized areas
- Areas with vegetation
- bare soils
- Water areas
In order this time to produce a map with nomenclature level 2 (more deatailed), the class bare soils includes areas/pixels that remain bare ground all year. And others are bare soil for the dry season and become a stratum of cleared vegetation for the wet season.How to discriminate those soils which are bare all year round and those which are bare and vegetated in the wet season ? How to for instance taken into consideration the ndvi/spectral temporal variation?
Ideas : Is it a good idea to spend time to exploit the file called Samples_region_1_seed0_learn ? In order to define the temporal ndvi thresholds ? Or Do you advise me to test other classification algorithms? (deep learning module?). I found this ressource dealing with the Temporal Convolutional Neural Network for the Classification of Satellite Image Time Series : text It would be more relevant than the RF ? If yes, is it available trhough iota2?
I share below my level 1 nomenclature result classif and FYI my my config file (rf algo for the classif).
chain :
{
output_path : '/home/geoteca/ie_tld/LABEX/iota2_Test/Dakar_Test5/IOTA2_Outputs/Results/'
remove_output_path : True
nomenclature_path : '/home/geoteca/ie_tld/LABEX/iota2_Test/Dakar_Test5/nomenclature.txt'
list_tile : 'T28PBB'
s2_s2c_path : '/home/geoteca/ie_tld/LABEX/iota2_Test/Dakar_Test5/sensor_data/'
ground_truth : '/home/geoteca/ie_tld/LABEX/iota2_Test/Dakar_Test5/vector_data/rd_t3.shp'
data_field : 'niv1_id'
spatial_resolution : 10
color_table : '/home/geoteca/ie_tld/LABEX/iota2_Test/Dakar_Test5/colorFile.txt'
proj : 'EPSG:32628'
first_step: "init"
last_step: "validation"
}
arg_train :
{
classifier : 'rf'
otb_classifier_options : {'classifier.rf.min': 5,'classifier.rf.max': 25 }
}
arg_classification :
{
classif_mode : 'separate'
}
task_retry_limits:
{
allowed_retry : 0
maximum_ram : 60.0
maximum_cpu : 12
}