Classification maps are now returned as int16.

Signed-off-by: Daniel Scheffler <danschef@gfz-potsdam.de>

gms_preprocessing/algorithms/classification.py

@@ -15,8 +15,8 @@ from geoarray import GeoArray
 from py_tools_ds.numeric.array import get_array_tilebounds
 
 
-global_shared_endmembers = None
-global_shared_im2classify = None
+global_shared_endmembers = None  # type: Union[None, np.ndarray]
+global_shared_im2classify = None  # type: Union[None, GeoArray]
 
 
 def initializer(endmembers, im2classify):
@@ -59,12 +59,11 @@ class _ImageClassifier(object):
         """
         self._cmap_nodataVal = cmap_nodataVal
 
-        dtype_cmap = np.array(self.train_labels).dtype
+        dtype_cmap = np.int16
         if cmap_nodataVal is not None and not np.can_cast(cmap_nodataVal, dtype_cmap):
             dtype_cmap = np.find_common_type(np.array(self.train_labels), np.array([cmap_nodataVal]))
 
-        image_cube_gA = GeoArray(image_cube, nodata=in_nodataVal)
-        # image_cube_gA.to_mem()
+        image_cube_gA = GeoArray(image_cube, nodata=in_nodataVal)  # lazily read in tiles to save memory
 
         bounds_alltiles = get_array_tilebounds(image_cube_gA.shape, tiledims)
 
@@ -222,7 +221,7 @@ class MinimumDistance_Classifier(_ImageClassifier):
 
         dist = self.compute_euclidian_distance(tileimdata.astype(np.float32), cmap, self._cmap_nodataVal)
 
-        return tilepos, cmap, dist
+        return tilepos, cmap.astype(np.int16), dist
 
     def label_unclassified_pixels(self, label_unclassified, threshold):
         # type: (int, Union[str, int, float]) -> GeoArray
@@ -253,7 +252,9 @@ class kNN_Classifier(_ImageClassifier):
         tileimdata = global_shared_im2classify[rS: rE + 1, cS: cE + 1, :]
 
         spectra = tileimdata.reshape((tileimdata.shape[0] * tileimdata.shape[1], tileimdata.shape[2]))
-        return tilepos, self.clf.predict(spectra).reshape(*tileimdata.shape[:2]), None
+        cmap = self.clf.predict(spectra).reshape(*tileimdata.shape[:2])
+
+        return tilepos, cmap.astype(np.int16), None
 
 
 class SAM_Classifier(_ImageClassifier):
@@ -276,7 +277,7 @@ class SAM_Classifier(_ImageClassifier):
 
         (rS, rE), (cS, cE) = tilepos
         tileimdata = global_shared_im2classify[rS: rE + 1, cS: cE + 1, :]
-        endmembers = global_shared_endmembers  # type: np.ndarray
+        endmembers = global_shared_endmembers
 
         if not tileimdata.shape[2] == self.train_spectra.shape[1]:
             raise RuntimeError('Matrix dimensions are not aligned. Input image has %d bands but input spectra '
@@ -302,7 +303,7 @@ class SAM_Classifier(_ImageClassifier):
             cmap = self.overwrite_cmap_at_nodata_positions(cmap, tileimdata,
                                                            self._cmap_nodataVal, global_shared_im2classify.nodata)
 
-        return tilepos, cmap, angles_min
+        return tilepos, cmap.astype(np.int16), angles_min
 
     @staticmethod
     def _calc_sam(s1_norm, s2_norm, axis=0):
@@ -349,7 +350,7 @@ class SID_Classifier(_ImageClassifier):
 
         (rS, rE), (cS, cE) = tilepos
         tileimdata = global_shared_im2classify[rS: rE + 1, cS: cE + 1, :]
-        endmembers = global_shared_endmembers  # type: np.ndarray
+        endmembers = global_shared_endmembers
 
         if not tileimdata.shape[2] == self.train_spectra.shape[1]:
             raise RuntimeError('Matrix dimensions are not aligned. Input image has %d bands but input spectra '
@@ -375,7 +376,7 @@ class SID_Classifier(_ImageClassifier):
             cmap = self.overwrite_cmap_at_nodata_positions(cmap, tileimdata,
                                                            self._cmap_nodataVal, global_shared_im2classify.nodata)
 
-        return tilepos, cmap, sid_min
+        return tilepos, cmap.astype(np.int16), sid_min
 
     @staticmethod
     def _calc_sid(s1_norm, s2_norm, axis=0):
@@ -431,7 +432,7 @@ class RF_Classifier(_ImageClassifier):
             cmap = self.overwrite_cmap_at_nodata_positions(cmap, tileimdata,
                                                            self._cmap_nodataVal, global_shared_im2classify.nodata)
 
-        return tilepos, cmap, None
+        return tilepos, cmap.astype(np.int16), None
 
 
 def classify_image(image, train_spectra, train_labels, classif_alg, in_nodataVal=None, cmap_nodataVal=None,