Num_Each_File[times - 1] = Num_Each_File[times - 1] + res_num
start = 0
prob_map = np.zeros((1, n_classes))
for i in range(times):
feed_x = np.reshape(
np.asarray(All_data['patch'][start:start + Num_Each_File[i]]),
(-1, n_input))
temp = sess.run(softmax_output, feed_dict={x: feed_x})
prob_map = np.concatenate((prob_map, temp), axis=0)
start += Num_Each_File[i]
prob_map = np.delete(prob_map, (0), axis=0)
# MRF
prob_map = compute_prob_map()
Seg_Label, seg_Label, seg_accuracy = Post_Processing(
prob_map, Height, Width, n_classes, y_test_scalar, TestIndex)
print('The shape of prob_map is (%d,%d)' %
(prob_map.shape[0], prob_map.shape[1]))
DATA_PATH = os.getcwd()
file_name = 'prob_map.mat'
prob = {}
prob['prob_map'] = prob_map
scipy.io.savemat(os.path.join(DATA_PATH, file_name), prob)
train_ind = {}
train_ind['TrainIndex'] = TrainIndex
scipy.io.savemat(os.path.join(DATA_PATH, 'TrainIndex.mat'), train_ind)
test_ind = {}
test_ind['TestIndex'] = TestIndex
test_map = test_map.reshape(GT_Label.shape[1],
GT_Label.shape[0]).transpose(1, 0).astype(int)
DATA_PATH = os.getcwd()
train_ind = {}
train_ind['train_indexes'] = train_indexes
scipy.io.savemat(os.path.join(DATA_PATH, 'train_indexes.mat'), train_ind)
test_ind = {}
test_ind['test_indexes'] = test_indexes
scipy.io.savemat(os.path.join(DATA_PATH, 'test_indexes.mat'), test_ind)
## Data Summary
df = data_summary(y_train, y, num_classes)
print('----------------------------------')
print('Data Summary:')
print(df)
print('----------------------------------')
print("Training samples: %d" % len(y_train))
print("Test samples: %d" % len(y_test))
print('----------------------------------')
DATA_PATH = os.path.join(os.getcwd(), "datasets")
prob_map = scipy.io.loadmat(os.path.join(DATA_PATH, 'p.mat'))['p']
prob_map = np.transpose(prob_map)
# Post-processing using Graph-Cut
Seg_Label, seg_accuracy = Post_Processing(prob_map,height,width,\
num_classes,y_test,test_indexes)
print(seg_accuracy)
scaler = MinMaxScaler()
X_train_scaled = scaler.fit_transform(X_train)
X_test_scaled = scaler.transform(X_test)
data_all_scaled = scaler.transform(data_all)
## Classifiers
# KNN
from sklearn.neighbors import KNeighborsClassifier
start_time = time.time()
KNN = KNeighborsClassifier(n_neighbors=7).fit(X_train_scaled, y_train)
KNN_Label = KNN.predict(data_all_scaled).reshape(width,
height).astype(int).transpose(
1, 0)
KNN_predict_prob = KNN.predict_proba(data_all_scaled)
# Post-processing using Graph-Cut
Seg_Label, seg_accuracy = Post_Processing(KNN_predict_prob,height,width,\
num_classes,y_test,test_indexes)
print('(KNN) Train_Acc=%.3f, Cla_Acc=%.3f, Seg_Acc=%.3f(Time_cost=%.3f)'\
% (KNN.score(X_train_scaled,y_train),KNN.score(X_test_scaled,y_test),\
seg_accuracy, (time.time()-start_time)))
# draw classification map
draw(GT_Label, KNN_Label, Seg_Label, train_map, test_map)
print('--------------------------------------------------------------------')
# Naive Bayes: GaussianNB
from sklearn.naive_bayes import GaussianNB
start_time = time.time()
GaussNB = GaussianNB().fit(X_train, y_train)
GaussNB_Label = GaussNB.predict(data_all).reshape(
width, height).astype(int).transpose(1, 0)
GaussNB_predict_prob = GaussNB.predict_proba(data_all)
# Post-processing using Graph-Cut