Skip to content

Commit 8c72324

Browse files
DotWangDotWang
authored
Add files via upload
1 parent 218d448 commit 8c72324

File tree

6 files changed

+1488
-0
lines changed

6 files changed

+1488
-0
lines changed

‎func.py

Lines changed: 241 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,241 @@
1+
import numpy as np
2+
import scipy.io as scio
3+
from sklearn.preprocessing import MinMaxScaler
4+
from sklearn.decomposition import FastICA
5+
from sklearn.decomposition import PCA
6+
import cv2
7+
8+
class load():
9+
# load dataset(indian_pines & pavia_univ.)
10+
def load_data(self,flag='indian'):
11+
if flag == 'indian':
12+
Ind_pines_dict = scio.loadmat('/data/di.wang/ordinary/23DCNN/Indian_pines.mat')
13+
Ind_pines_gt_dict = scio.loadmat('/data/di.wang/ordinary/23DCNN/Indian_pines_gt.mat')
14+
15+
print(Ind_pines_dict['indian_pines'].shape)
16+
print(Ind_pines_gt_dict['indian_pines_gt'].shape)
17+
18+
# remove the water absorption bands
19+
20+
no_absorption = list(set(np.arange(0, 103)) | set(np.arange(108, 149)) | set(np.arange(163, 219)))
21+
22+
original = Ind_pines_dict['indian_pines'][:, :, no_absorption].reshape(145 * 145, 200)
23+
24+
print(original.shape)
25+
print('Remove wate absorption bands successfully!')
26+
27+
gt = Ind_pines_gt_dict['indian_pines_gt'].reshape(145 * 145, 1)
28+
29+
r = Ind_pines_dict['indian_pines'].shape[0]
30+
c = Ind_pines_dict['indian_pines'].shape[1]
31+
categories = 17
32+
if flag == 'pavia':
33+
pav_univ_dict = scio.loadmat('/data/di.wang/ordinary/23DCNN/PaviaU.mat')
34+
pav_univ_gt_dict = scio.loadmat('/data/di.wang/ordinary/23DCNN/PaviaU_gt.mat')
35+
36+
print(pav_univ_dict['paviaU'].shape)
37+
print(pav_univ_gt_dict['paviaU_gt'].shape)
38+
39+
original = pav_univ_dict['paviaU'].reshape(610 * 340, 103)
40+
gt = pav_univ_gt_dict['paviaU_gt'].reshape(610 * 340, 1)
41+
42+
r = pav_univ_dict['paviaU'].shape[0]
43+
c = pav_univ_dict['paviaU'].shape[1]
44+
categories = 10
45+
if flag == 'ksc':
46+
ksc_dict = scio.loadmat('/data/di.wang/ordinary/23DCNN/KSC.mat')
47+
ksc_gt_dict=scio.loadmat('/data/di.wang/ordinary/23DCNN/KSC_gt.mat')
48+
49+
print(ksc_dict['KSC'].shape)
50+
print(ksc_gt_dict['KSC_gt'].shape)
51+
52+
original = ksc_dict['KSC'].reshape(512 * 614, 176)
53+
original[original>400]=0
54+
gt = ksc_gt_dict['KSC_gt'].reshape(512 * 614, 1)
55+
56+
r = ksc_dict['KSC'].shape[0]
57+
c = ksc_dict['KSC'].shape[1]
58+
categories = 14
59+
60+
rows = np.arange(gt.shape[0]) # start from 0
61+
# ID(row number), data, class number
62+
All_data = np.c_[rows, original, gt]
63+
64+
# Removing background and obtain all labeled data
65+
labeled_data = All_data[All_data[:, -1] != 0, :]
66+
rows_num = labeled_data[:, 0] # All ID of labeled data
67+
68+
return All_data, labeled_data, rows_num, categories, r, c, flag
69+
70+
71+
class product():
72+
def __init__(self,c,flag):
73+
self.c=c
74+
self.flag=flag
75+
# product the training and testing pixel ID
76+
def generation_num(self,labeled_data, rows_num, All_data):
77+
78+
train_num = []
79+
80+
for i in np.unique(labeled_data[:, -1]):
81+
temp = labeled_data[labeled_data[:, -1] == i, :]
82+
temp_num = temp[:, 0] # all ID of a special class
83+
#print(i, temp_num.shape[0])
84+
np.random.shuffle(temp_num) # random sequence
85+
if self.flag == 'indian':
86+
if i == 1:
87+
train_num.append(temp_num[0:33])
88+
elif i == 7:
89+
train_num.append(temp_num[0:20])
90+
elif i == 9:
91+
train_num.append(temp_num[0:14])
92+
elif i == 16:
93+
train_num.append(temp_num[0:75])
94+
else:
95+
train_num.append(temp_num[0:100])
96+
if self.flag == 'pavia':
97+
train_num.append(temp_num[0:100])
98+
if self.flag == 'ksc':
99+
if i==1:
100+
train_num.append(temp_num[0:33])
101+
elif i==2:
102+
train_num.append(temp_num[0:23])
103+
elif i==3:
104+
train_num.append(temp_num[0:24])
105+
elif i==4:
106+
train_num.append(temp_num[0:24])
107+
elif i==5:
108+
train_num.append(temp_num[0:15])
109+
elif i==6:
110+
train_num.append(temp_num[0:22])
111+
elif i==7:
112+
train_num.append(temp_num[0:9])
113+
elif i==8:
114+
train_num.append(temp_num[0:38])
115+
elif i==9:
116+
train_num.append(temp_num[0:51])
117+
elif i==10:
118+
train_num.append(temp_num[0:39])
119+
elif i==11:
120+
train_num.append(temp_num[0:41])
121+
elif i==12:
122+
train_num.append(temp_num[0:49])
123+
elif i==13:
124+
train_num.append(temp_num[0:91])
125+
# else:
126+
# train_num.append(temp_num[0:int(temp.shape[0]*0.1)])
127+
128+
trn_num = [x for j in train_num for x in j] # merge
129+
tes_num = list(set(rows_num) - set(trn_num))
130+
pre_num = list(set(range(0, All_data.shape[0])) - set(trn_num))
131+
print('number of training sample', len(trn_num))
132+
return rows_num, trn_num, tes_num, pre_num
133+
134+
135+
def production_data_trn(self, rows_num, trn_num, half_s, image_3d_mat):
136+
137+
trn_num = np.array(trn_num)
138+
##Training set(spatial)
139+
idx_2d_trn = np.zeros([trn_num.shape[0], 2]).astype(int)
140+
idx_2d_trn[:, 0] = np.floor(trn_num / self.c)
141+
idx_2d_trn[:, 1] = trn_num + 1 - self.c * idx_2d_trn[:, 0] - 1
142+
# neibour area(2*half_s+1)
143+
patch_size=2*half_s+1
144+
trn_spat = np.zeros([trn_num.shape[0], patch_size, patch_size, image_3d_mat.shape[2]])
145+
neibour_num = []
146+
for i in range(idx_2d_trn.shape[0]):
147+
# image expandision
148+
row = idx_2d_trn[i, 0] + half_s
149+
col = idx_2d_trn[i, 1] + half_s
150+
trn_spat[i, :, :, :] = image_3d_mat[(row - half_s):row + half_s + 1,
151+
(col - half_s):col + half_s + 1, :]
152+
# mapping expandision neibor pixel ID to origianal image
153+
neibour_num = neibour_num + [(row + j - half_s) * self.c + col + k - half_s for j in range(-half_s, half_s+1) for k
154+
in range(-half_s, half_s+1)]
155+
val_num = list(set(rows_num) - set(neibour_num)) # prevent data snooping
156+
157+
print('trn_spat:', trn_spat.shape)
158+
print('Training Spatial dataset preparation Finished!')
159+
return trn_spat, trn_num, val_num
160+
161+
def production_data_valtespre(self, tes_num, half_s, image_3d_mat, flag='Tes'):
162+
163+
##Testing set(spatial)
164+
tes_num = np.array(tes_num)
165+
idx_2d_tes = np.zeros([tes_num.shape[0], 2]).astype(int)
166+
idx_2d_tes[:, 0] = np.floor(tes_num / self.c)
167+
idx_2d_tes[:, 1] = tes_num + 1 - self.c * idx_2d_tes[:, 0] - 1
168+
# neibour area(2*half_s+1)
169+
patch_size = 2 * half_s + 1
170+
tes_spat = np.zeros([tes_num.shape[0], patch_size, patch_size, image_3d_mat.shape[2]])
171+
for i in range(idx_2d_tes.shape[0]):
172+
# image expandision
173+
row = idx_2d_tes[i, 0] + half_s
174+
col = idx_2d_tes[i, 1] + half_s
175+
tes_spat[i, :, :, :] = image_3d_mat[(row - half_s):row + half_s + 1,
176+
(col - half_s):col + half_s + 1, :]
177+
178+
print('tes_spat:', tes_spat.shape)
179+
print('{} Spatial dataset preparation Finished!'.format(flag))
180+
return tes_spat,tes_num
181+
182+
def normlization(self, data_spat, mi, ma,flag='trn'):
183+
184+
scaler = MinMaxScaler(feature_range=(mi, ma))
185+
186+
spat_data = data_spat.reshape(-1, data_spat.shape[-1])
187+
data_spat_new = scaler.fit_transform(spat_data).reshape(data_spat.shape)
188+
189+
print('{}_spat:{}'.format(flag,data_spat_new.shape))
190+
print('{} Spatial dataset normalization Finished!'.format(flag))
191+
return data_spat_new
192+
193+
def resample(self,data,rsz):
194+
195+
if data.shape[1]==rsz:
196+
return data
197+
else:
198+
# data:BHWC
199+
data = data.transpose(0,3,1,2)#BCHW
200+
B,C,H,W=data.shape
201+
re_data = np.zeros([B,C,rsz,rsz])
202+
for i in range(B):
203+
for j in range(C):
204+
temp=data[i,j,:,:]
205+
re_data[i,j,:,:] = cv2.resize(temp, (rsz, rsz), interpolation=cv2.INTER_LINEAR)
206+
re_data=re_data.transpose(0,2,3,1)#BHWC
207+
return re_data
208+
209+
210+
211+
class preprocess():
212+
def __init__(self,t,dr_num):
213+
self.transform=t
214+
self.dr_num=dr_num
215+
def Dim_reduction(self, All_data):
216+
217+
Alldata_DR=All_data
218+
219+
# if self.transform =='ica':
220+
# ica_data_pre = All_data[:, 1:-1]
221+
# print(ica_data_pre.shape)
222+
# transformer = FastICA(n_components=50, whiten=True, random_state=None)
223+
# fastica_data = transformer.fit_transform(ica_data_pre)
224+
# print(fastica_data.shape)
225+
#
226+
# Alldata_DR = fastica_data
227+
#
228+
# print('ICA Finished!')
229+
230+
if self.transform =='pca':
231+
pca_data_pre = All_data[:, 1:-1]
232+
print(pca_data_pre.shape)
233+
pca_transformer = PCA(n_components=self.dr_num)
234+
pca_data = pca_transformer.fit_transform(All_data[:, 1:-1])
235+
print(pca_data.shape)
236+
237+
Alldata_DR = pca_data
238+
239+
print('PCA Finished!')
240+
241+
return Alldata_DR

0 commit comments

Comments
 (0)