dataTrain = data.Iterator(
dataSamples=hypData.spectraPrep[:trainSamples, :],
targets=hypData.spectraPrep[:trainSamples, :],
batchSize=1000)
dataVal = data.Iterator(
dataSamples=hypData.spectraPrep[trainSamples:trainSamples +
valSamples, :],
targets=hypData.spectraPrep[trainSamples:trainSamples + valSamples, :])
# shuffle training data
dataTrain.shuffle()
# setup a fully-connected autoencoder neural network with 3 encoder layers
net_mlp = autoencoder.mlp_1D_network(inputSize=hypData.numBands,
encoderSize=[50, 30, 10, 3],
activationFunc='relu',
weightInitOpt='truncated_normal',
tiedWeights=None,
skipConnect=False)
# setup a convolutional autoencoder neural network with 3 conv encoder layers
net_cnn = autoencoder.cnn_1D_network(inputSize=hypData.numBands,
zDim=3,
encoderNumFilters=[10, 10, 10],
encoderFilterSize=[20, 10, 10],
activationFunc='relu',
weightInitOpt='truncated_normal',
encoderStride=[1, 1, 1],
tiedWeights=None,
skipConnect=False)
# setup a training operation for each network (using the same loss function)
from deephyp import data
if __name__ == '__main__':
# read data into numpy array
mat = scipy.io.loadmat('PaviaU.mat')
img = mat['paviaU']
# create a hyperspectral dataset object from the numpy array
hypData = data.HypImg(img)
# pre-process data to make the model easier to train
hypData.pre_process('minmax')
# setup each network from the config files
net_mlp = autoencoder.mlp_1D_network(configFile=os.path.join(
'models', 'test_ae_comparison_mlp', 'config.json'))
net_cnn = autoencoder.cnn_1D_network(configFile=os.path.join(
'models', 'test_ae_comparison_cnn', 'config.json'))
# assign previously trained parameters to the network, and name each model
net_mlp.add_model(addr=os.path.join('models', 'test_ae_comparison_mlp',
'epoch_100'),
modelName='mlp_100')
net_cnn.add_model(addr=os.path.join('models', 'test_ae_comparison_cnn',
'epoch_10'),
modelName='cnn_10')
# feed forward hyperspectral dataset through each encoder model (get latent encoding)
dataZ_mlp = net_mlp.encoder(modelName='mlp_100',
dataSamples=hypData.spectraPrep)
dataZ_cnn = net_cnn.encoder(modelName='cnn_10',
from deephyp import data
if __name__ == '__main__':
# read data into numpy array
mat = scipy.io.loadmat('PaviaU.mat')
img = mat['paviaU']
# create a hyperspectral dataset object from the numpy array
hypData = data.HypImg(img)
# pre-process data to make the model easier to train
hypData.pre_process('minmax')
# setup a network from a config file
net = autoencoder.mlp_1D_network(
configFile=os.path.join('models', 'test_ae_mlp_sid', 'config.json'))
# assign previously trained parameters to the network, and name model
net.add_model(addr=os.path.join('models', 'test_ae_mlp_sid', 'epoch_100'),
modelName='sid_100')
# feed forward hyperspectral dataset through encoder (get latent encoding)
dataZ = net.encoder(modelName='sid_100', dataSamples=hypData.spectraPrep)
# feed forward latent encoding through decoder (get reconstruction)
dataY = net.decoder(modelName='sid_100', dataZ=dataZ)
#--------- visualisation ----------------------------------------
# download dataset ground truth pixel labels (if already downloaded, comment this out)
urlretrieve('http://www.ehu.eus/ccwintco/uploads/5/50/PaviaU_gt.mat',
hypData.pre_process( 'minmax' )
# create data iterator objects for training and validation using the pre-processed data
trainSamples = 200000
valSamples = 100
dataTrain = data.Iterator( dataSamples=hypData.spectraPrep[:trainSamples, :],
targets=hypData.spectraPrep[:trainSamples, :], batchSize=1000 )
dataVal = data.Iterator( dataSamples=hypData.spectraPrep[trainSamples:trainSamples+valSamples, :],
targets=hypData.spectraPrep[trainSamples:trainSamples+valSamples, :] )
# shuffle training data
dataTrain.shuffle()
# setup a fully-connected autoencoder neural network with 3 encoder layers
net = autoencoder.mlp_1D_network( inputSize=hypData.numBands, encoderSize=[50,30,10], activationFunc='sigmoid',
weightInitOpt='truncated_normal', tiedWeights=None, skipConnect=False,
activationFuncFinal='sigmoid' )
# setup a training operation for the network
net.add_train_op( name='sid', lossFunc='SID', learning_rate=1e-3, decay_steps=None, decay_rate=None,
method='Adam', wd_lambda=0.0 )
# create a directory to save the learnt model
model_dir = os.path.join('models','test_ae_mlp_sid')
if os.path.exists(model_dir):
# if directory already exists, delete it
shutil.rmtree(model_dir)
os.mkdir(model_dir)
# train the network for 100 epochs, saving the model at epoch 50 and 100
net.train(dataTrain=dataTrain, dataVal=dataVal, train_op_name='sid', n_epochs=100, save_addr=model_dir,
