diff --git a/Learning/ObjectBased.py b/Learning/ObjectBased.py
index 93effce4b23119aa2320ea42a11d115249a340ee..3d23e444f615cfd3f50cd104ee922651172c7863 100644
--- a/Learning/ObjectBased.py
+++ b/Learning/ObjectBased.py
@@ -9,17 +9,18 @@ from sklearn.ensemble import RandomForestClassifier
from sklearn.metrics import confusion_matrix, accuracy_score, cohen_kappa_score, precision_recall_fscore_support
class ObjectBasedClassifier:
- def __init__(self, object_layer, reference_data, time_series_list, user_feature_list, ref_class_field='class',
- ref_id_field='id'):
+ def __init__(self, object_layer, time_series_list, user_feature_list,
+ reference_data=None, ref_class_field='class', ref_id_field='id'):
self.obia_base = OBIABase(object_layer, ref_data=reference_data, ref_class_field=ref_class_field,
ref_id_field=ref_id_field)
for lst in time_series_list:
self.obia_base.add_raster_time_series_for_stats(lst)
for ras in user_feature_list:
self.obia_base.add_raster_for_stats(ras)
- self.obia_base.populate_ref_db()
- self.training_base = self.obia_base.get_reference_db_as_training_base(class_field=ref_class_field)
- self.training_base['folds'] = []
+ if reference_data is not None:
+ self.obia_base.populate_ref_db()
+ self.training_base = self.obia_base.get_reference_db_as_training_base(class_field=ref_class_field)
+ self.training_base['folds'] = []
return
def gen_k_folds(self, k, class_field='class'):
@@ -96,11 +97,10 @@ class ObjectBasedClassifier:
}
return models, summary, results
- def classify(self, model, output_file=None, compress='NONE'):
+ def classify(self, model, perc=None, output_file=None, compress='NONE'):
prg = tqdm(desc='Classification', total=len(self.obia_base.tiles))
if isinstance(model, list):
- for t, L, X in self.obia_base.tiled_data(normalize=[self.training_base['perc2'],
- self.training_base['perc98']]):
+ for t, L, X in self.obia_base.tiled_data(normalize=perc):
prob = []
for m in model:
prob.append(m.predict_proba(X))
@@ -109,8 +109,7 @@ class ObjectBasedClassifier:
self.obia_base.populate_map(t, L, c, output_file, compress)
prg.update(1)
else:
- for t,L,X in self.obia_base.tiled_data(normalize=[self.training_base['perc2'],
- self.training_base['perc98']]):
+ for t,L,X in self.obia_base.tiled_data(normalize=perc):
c = model.predict(X)
self.obia_base.populate_map(t, L, c, output_file, compress)
prg.update(1)
diff --git a/Workflows/basic.py b/Workflows/basic.py
index 2fd290506f95550ef004ebbd56ca53fdfc82de6d..e8191481278ddffec8f85156d6e25c89af5dff7a 100644
--- a/Workflows/basic.py
+++ b/Workflows/basic.py
@@ -3,6 +3,7 @@ import json
import pickle
from moringa import preprocess_s2, run_segmentation
from Learning.ObjectBased import ObjectBasedClassifier
+from Postprocessing import Report
def process_timeseries(oroot, d, ts_lst_pkl):
ts_lst = []
@@ -44,20 +45,58 @@ def train_valid_workflow(seg, ts_lst_pkl, d, m_file):
with open(ts_lst_pkl, 'rb') as ts_save:
ts_lst = pickle.load(ts_save)
obc = ObjectBasedClassifier(seg,
- d['ref_db']['path'],
ts_lst,
d['userfeat'],
+ reference_data=d['ref_db']['path'],
ref_class_field=d['ref_db']['fields'])
obc.gen_k_folds(5, class_field=d['ref_db']['fields'][-1])
- for cf in d['ref_db']['fields']:
+ for i,cf in enumerate(d['ref_db']['fields']):
if d['training']['classifier'] == 'rf':
m, s, r = obc.train_RF(d['training']['parameters']['n_trees'], class_field=cf, return_true_vs_pred=True)
- m_dict = {'model': m, 'results': r, 'summary': s}
- os.makedirs(m_file[cf], exist_ok=True)
- with open(m_file[cf], 'wb') as mf:
+ m_dict = {'model': m, 'results': r, 'summary': s,
+ 'perc2':obc.training_base['perc2'], 'perc98':obc.training_base['perc98']}
+ os.makedirs(os.path.dirname(m_file[i]), exist_ok=True)
+ with open(m_file[i], 'wb') as mf:
pickle.dump(m_dict, mf)
+ return
+
+def classify(seg, ts_lst_pkl, m_files, d, map_files):
+ assert (os.path.exists(seg))
+ assert (os.path.exists(ts_lst_pkl))
+ for m_file in m_files:
+ assert (os.path.exists(m_file))
+ print('[MORINGA-INFO] : Performing classification')
+ with open(ts_lst_pkl, 'rb') as ts_save:
+ ts_lst = pickle.load(ts_save)
+ obc = ObjectBasedClassifier(seg,
+ ts_lst,
+ d['userfeat'])
+ for m_file, map_file in zip(m_files, map_files):
+ with open(m_file, 'rb') as mf:
+ m_dict = pickle.load(mf)
+ obc.classify(m_dict['model'], perc=[m_dict['perc2'], m_dict['perc98']], output_file=map_file)
+ return
+
+def report(map_files, m_files, d, report_files):
+ print('[MORINGA-INFO] : Generating report(s)')
+ for map_file, palette_fn, m_file, report_file in zip(map_files, d['map_output']['palette_files'], m_files, report_files):
+ assert os.path.exists(map_file)
+ assert os.path.exists(m_file)
+ os.makedirs(os.path.splitext(report_file)[0]+'_figures', exist_ok=True)
+ with open(m_file, 'rb') as mf:
+ m_dict = pickle.load(mf)
+ of = Report.generate_report_figures(
+ map_file,
+ palette_fn,
+ m_dict['results'],
+ m_dict['summary'],
+ os.path.splitext(report_file)[0]+'_figures',
+ d['chain_name'])
+ Report.generate_pdf(of, report_file, d['chain_name'])
+ return
+
def basic(cfg, runlevel=1, single_step=False):
os.environ['OTB_LOGGER_LEVEL'] = 'CRITICAL'
with open(cfg,'r') as f:
@@ -84,15 +123,32 @@ def basic(cfg, runlevel=1, single_step=False):
return
# Training/Validation Workflow
- m_file = {}
+ m_files = []
for cf in d['ref_db']['fields']:
- m_file[cf] = os.path.join(oroot, 'model/model_{}.pkl'.format(cf))
+ m_files.append(os.path.join(oroot, 'model/model_{}.pkl'.format(cf)))
if step == 3:
- train_valid_workflow(seg, ts_lst_pkl, d, m_file)
+ train_valid_workflow(seg, ts_lst_pkl, d, m_files)
+ step += 1
+ if single_step:
+ return
+
+ # Classification
+ map_files = []
+ for cf in d['ref_db']['fields']:
+ map_files.append(os.path.join(oroot, 'maps/{}_map_{}.tif'.format(d['chain_name'],cf)))
+ if step == 4:
+ classify(seg, ts_lst_pkl, m_files, d, map_files)
step += 1
if single_step:
return
+ # Report
+ report_fn = []
+ for cf in d['ref_db']['fields']:
+ report_fn.append(os.path.join(oroot, 'reports/{}_report_{}.pdf'.format(d['chain_name'],cf)))
+ if step == 5:
+ report(map_files, m_files, d, report_fn)
+
return