-
Notifications
You must be signed in to change notification settings - Fork 29
/
Copy pathfuncCNN.py
205 lines (184 loc) · 6.76 KB
/
funcCNN.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
import numpy as np
import scipy.io as scio
import tensorflow as tf
def Con2Numpy(var_name):
dataFile = var_name
data = scio.loadmat(dataFile)
x=data[var_name]
x1=x.astype(float)
return x1
def Con2Numpy_path(var_name, path):
dataFile = var_name
data = scio.loadmat(path+dataFile)
x=data[var_name]
x1=x.astype(float)
return x1
def load_HSI_data_list(path1):
inst = np.array(Con2Numpy_path('inst', path1), dtype = 'float32')
temask= np.array(Con2Numpy_path('temask', path1), dtype=bool)
trmask=np.array(Con2Numpy_path('trmask', path1), dtype=bool)
y_train=Con2Numpy_path('y_train', path1)
y_test=Con2Numpy_path('y_test', path1)
return inst, temask, trmask, y_train, y_test
def load_HSI_data():
features_pretrain = np.array(Con2Numpy('features_pretrain'), dtype = 'float32')
support_pretrain = np.array(Con2Numpy('support_pretrain'), dtype = 'float32')
trmask_pretrain = Con2Numpy('trmask_pretrain')
trmask_pretrain = np.array(trmask_pretrain, dtype=bool)
y_train_pretrain = Con2Numpy('y_train_pretrain')
return features_pretrain, support_pretrain, trmask_pretrain, y_train_pretrain
def AscSort(x1):
x = x1.copy()
B = np.sort(x)
IX = np.ones(np.size(x))
for i in range(np.size(x)):
idx = np.argmin(x)
x[idx] = np.max(x)+1
IX[i] = idx
return B, IX
def GetKnn1(inx, data, k):
[datarow, datacol] = np.shape(data)
diffMat = np.tile(inx, (datarow, 1)) - data
distanceMat = np.sqrt(np.sum(diffMat*diffMat, 1))
[B, IX] = AscSort(distanceMat)
if B[0] == 0:
IX = IX[1:k+1].copy()
else:
IX = IX[0:k].copy()
return IX
def lle(X, K, ln):
[D, N] = np.shape(X)
index = np.zeros([K, N])
for i in range(N):
index[:, i] = GetKnn1(X[:, i].T, X[:, 0:ln].T, K)
neighborhood = index[0:K, :]
if K>D:
tol = 0.001
else:
tol = 0
W = np.zeros([K, N])
W1 = np.zeros([N, N])
for ii in range(N):
z = X[:, np.array(neighborhood[:, ii], dtype=int)]-np.tile(X[:, ii], (K, 1)).T
C = np.dot(z.T, z)
C = C + np.dot(np.dot(np.eye(K), tol), np.trace(C))
W[:, ii] = np.dot(np.linalg.inv(C), np.ones([K,1])).reshape((K))
W[:, ii] = W[:, ii]/np.sum(W[:, ii])
for i in range(N):
for j in range(K):
W1[np.array(index[j, i], dtype=int), i] = W[j, i]
return W1
def GetWlle(S1):
S = S1.copy()
S = S.T
Wlle = S+S.T-np.dot(S.T, S)
Wlle = Wlle-np.diag(np.diag(Wlle))
return Wlle
def GetLabeledData(img2d, trte_idx):
intrte=np.zeros([np.shape(trte_idx)[0],np.shape(img2d)[1]])
all_num=np.shape(trte_idx)[0]
for i in range(all_num):
print(i)
intrte[i,:] = img2d[trte_idx[i,0], :]
intrte=np.reshape(intrte, (all_num ,40))
return intrte
def GetMats(trte_idx,y_test,y_train,trnum):
nums=np.shape(trte_idx)[0]
trmask=np.zeros([nums])
temask=np.zeros([nums])
ytr=np.zeros([nums,np.shape(y_test)[1]])
yte=np.zeros([nums,np.shape(y_test)[1]])
for i in range(trnum):
trmask[i]=1
ytr[i]=y_train[trte_idx[i],:]
for i in range(nums-trnum):
temask[i+trnum]=1
yte[i+trnum]=y_test[trte_idx[i+trnum],:]
return np.array(trmask,dtype=bool),np.array(temask,dtype=bool),ytr,yte
def Gettrtemask(dim0,trnum,trte):
if trte==0:
trte_mask1=np.ones([trnum])
trte_mask2=np.zeros([dim0-trnum])
return np.concatenate([trte_mask1,trte_mask2],axis=0)
elif trte==1:
trte_mask1=np.zeros([trnum])
trte_mask2=np.ones([dim0-trnum])
return np.concatenate([trte_mask1,trte_mask2],axis=0)
def mapminmax01(X):
min_max_scaler = preprocessing.MinMaxScaler()
return min_max_scaler.fit_transform(X)
def CalSupport_tf(A, lam):
lam1 = lam
A_ = A+lam1*np.eye(np.shape(A)[0])
D_ = tf.reduce_sum(A_,reduction_indices=0)
D_05 = tf.diag(D_**(-0.5))
support = tf.matmul(tf.matmul(D_05, A_), D_05)
return support
def CalSupport(A, lam):
lam1 = lam
A_ = A+lam1*np.eye(np.shape(A)[0])
D_ = np.sum(A_, 1)
D_05 = np.diag(D_**(-0.5))
support = np.matmul(np.matmul(D_05, A_), D_05)
return support
def arr2sparse(arr):
arr_tensor = tf.constant(np.array(arr),dtype='float32')
arr_idx = tf.where(tf.not_equal(arr_tensor, 0))
arr_sparse = tf.SparseTensor(arr_idx, tf.gather_nd(arr_tensor, arr_idx), arr_tensor.get_shape())
return arr_sparse
def AssignLabels(useful_sp_lab, trlabels, telabels, trmask, temask):
[rows, cols] = np.shape(useful_sp_lab)
output_labels = np.zeros([rows, cols])
sp_num = np.max(useful_sp_lab)
for sp_idx in range(1, sp_num+1):
pos1 = np.argwhere(useful_sp_lab==sp_idx)
if trmask[sp_idx-1, 0] == True:
pred_label = trlabels[sp_idx-1]
else:
pred_label = telabels[sp_idx-1]
output_labels[pos1[:,0], pos1[:,1]] = pred_label+1
return output_labels
def PixelWiseAccuracy(gt, pred_labels, trpos):
num_labels = np.max(gt)
gt[trpos[:,0]-1, trpos[:,1]-1] = 0
err_num = 0
for label_idx in range(1, num_labels+1):
pos1 = np.argwhere(gt == label_idx)
mat1 = gt[pos1[:,0], pos1[:,1]]
mat2 = pred_labels[pos1[:,0], pos1[:,1]]
mat3 = mat1-mat2
err_num += np.shape(np.argwhere(mat3!=0))[0]
return 1-err_num/np.shape(np.argwhere(gt>0))[0]
def GetExcelData(gt, pred_labels, trpos):
gt[trpos[:,0]-1, trpos[:,1]-1] = 0
num_classes = np.max(gt)
per_acc = []
overall_err_num = 0
for lab_idx in range(1, num_classes+1):
pos1 = np.argwhere(gt==lab_idx)
preds = pred_labels[pos1[:,0], pos1[:,1]]
gts = gt[pos1[:,0], pos1[:,1]]
mat3 = gts-preds
per_err_num = np.shape(np.argwhere(mat3!=0))[0]
per_acc.append(1-per_err_num/np.shape(pos1)[0])
overall_err_num += np.shape(np.argwhere(mat3!=0))[0]
per_acc = np.array(per_acc, dtype='float32')
OA = 1-overall_err_num/np.shape(np.argwhere(gt>0))[0]
AA = np.mean(per_acc)
# kappa
n = np.shape(np.argwhere(gt!=0))[0]
ab1 = 0
pos0 = np.argwhere(gt==0)
pred_labels[pos0[:,0], pos0[:,1]] = 0
for lab_idx in range(1, num_classes+1):
a1 = np.shape(np.argwhere(gt==lab_idx))[0]
b1 = np.shape(np.argwhere(pred_labels==lab_idx))[0]
ab1 += a1*b1
Pe = ab1/(n*n)
kappa_coef = (OA-Pe)/(1-Pe)
outputs = np.zeros([num_classes+3])
outputs[0:num_classes] = per_acc
outputs[num_classes] = OA
outputs[num_classes+1] = AA
outputs[num_classes+2] = kappa_coef
return outputs