def randomtree_lidaralldata17():
alltest_y = np.array([])
alltest_y.shape = 0
K.clear_session()
model = lidar_model()
model.load_weights('model/lidar_model17.h5')
print(model.summary())
model = Model(inputs=model.input,
outputs=model.get_layer('dense_2').output)
print(model.summary())
rf = joblib.load('model/rflidar17.model')
for m in range(8, 2412):
print str(m)
lidar_batch_patch = np.zeros(
(8344, patch_size_LID, patch_size_LID, lidar_channel),
dtype=np.uint8)
for n in range(8, 8352):
bias = int(patch_size_LID / 2)
upper_y = m - bias
left_x = n - bias
lidar_batch_patch[n -
8] = lidar_change[upper_y:upper_y +
patch_size_LID,
left_x:left_x + patch_size_LID]
pre_labels = model.predict(lidar_batch_patch)
pre_labels = pre_labels.reshape((8344, 512))
predicted_yone = rf.predict(pre_labels)
alltest_y = np.concatenate((alltest_y, predicted_yone), 0)
alltest_y = alltest_y.reshape((2404, 8344))
np.save("Result/alltest_y17.npy", alltest_y)
K.clear_session()
def randomtree_lidar3():
test_x = np.array([])
test_x.shape = 0, 512
test_y = np.array([])
test_y.shape = 0,
K.clear_session()
model = lidar_model()
model.load_weights('model/lidar_model3.h5')
print(model.summary())
model = Model(inputs=model.input,
outputs=model.get_layer('dense_2').output)
print(model.summary())
for j in range(int(label_location3[0].shape[0] // 1000)):
print str(j)
lidar_batch_label = np.zeros((batch_size), dtype=int)
lidar_batch_patch = np.zeros(
(batch_size, patch_size_LID, patch_size_LID, lidar_channel),
dtype=np.uint8)
for i in range(1000):
num = j * 1000 + i
X = label_location3[0][num]
Y = label_location3[1][num]
lab = train_sets3[X, Y]
lidar_batch_label[i] = lab
y = label_location3[0][num]
x = label_location3[1][num]
bias = int(patch_size_LID / 2)
upper_y = y - bias
left_x = x - bias
upper_y, left_x = boundary_judge(upper_y, left_x, patch_size_LID,
lidar.shape[0], lidar.shape[1])
lidar_batch_patch[i] = lidar[upper_y:upper_y + patch_size_LID,
left_x:left_x + patch_size_LID]
test_y = np.concatenate((test_y, lidar_batch_label), 0)
pre_labels = model.predict(lidar_batch_patch)
pre_labels = pre_labels.reshape((test_size, 512))
test_x = np.concatenate((test_x, pre_labels), 0)
K.clear_session()
rf = RandomForestClassifier(criterion="entropy",
max_features="sqrt",
n_estimators=400,
min_samples_leaf=2,
n_jobs=-1,
oob_score=True)
rf.fit(test_x, test_y)
joblib.dump(rf, 'model/rflidar3.model')
print('OK!')
def lidartrain17():
total_number = 25000
save_number = 1000
K.clear_session()
model = lidar_model()
t0 = time.time()
for i in range(total_number):
print "i " + str(i)
lidar, label = lidar_train_data17(batch_size=batch_size_LID, patch_size=patch_size_LID)
model.fit(lidar, label, epochs=5, batch_size=batch_size_LID)
if i % save_number == 0:
print "save model......"
model.save_weights('model/lidar_model17.h5')
K.clear_session()
t1 = time.time()
print "spend total time:" + str(round(t1 - t0, 2)) + "s"
K.clear_session()
def lidar_test():
K.clear_session()
model = lidar_model()
model.load_weights('model/lidartwo_model.h5')
label = np.array([])
label.shape = 0, 2
for m in range(8, 2412):
print str(m)
lidar_batch_patch = np.zeros(
(8344, patch_size_LID, patch_size_LID, lidar_channel),
dtype=np.uint8)
for n in range(8, 8352):
bias = int(patch_size_LID / 2)
upper_y = m - bias
left_x = n - bias
lidar_batch_patch[n - 8, ...] = lidar_change[upper_y:upper_y +
patch_size_LID,
left_x:left_x +
patch_size_LID, :]
pre_labels = model.predict(lidar_batch_patch)
label = np.concatenate((label, pre_labels), 0)
label = label.reshape(2404, 8344, 2)
np.save('Result/labelvggtwo.npy', label)
K.clear_session()