
!nvidia-smi

Sun May  3 22:00:11 2020       
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 440.64.00    Driver Version: 440.64.00    CUDA Version: 10.2     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  TITAN X (Pascal)    On   | 00000000:05:00.0 Off |                  N/A |
| 23%   38C    P8     9W / 250W |    114MiB / 12194MiB |      0%      Default |
+-------------------------------+----------------------+----------------------+
|   1  GeForce GTX 108...  On   | 00000000:09:00.0 Off |                  N/A |
| 23%   34C    P8     9W / 250W |      2MiB / 11178MiB |      0%      Default |
+-------------------------------+----------------------+----------------------+
|   2  TITAN RTX           On   | 00000000:0B:00.0 Off |                  N/A |
| 40%   39C    P0    84W / 280W |      1MiB / 24220MiB |     31%      Default |
+-------------------------------+----------------------+----------------------+
                                                                               
+-----------------------------------------------------------------------------+
| Processes:                                                       GPU Memory |
|  GPU       PID   Type   Process name                             Usage      |
|=============================================================================|
|    0      1596      G   /usr/lib/xorg/Xorg                            39MiB |
|    0      1633      G   /usr/bin/gnome-shell                          72MiB |
+-----------------------------------------------------------------------------+


# Imports para manipulação e visualização de dados
import os
import sys
import cv2
import csv
import math
import time
import pickle
import numpy as np
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
import matplotlib.image as mpimg
from os import listdir
from tqdm import tqdm
from PIL import Image
from mpl_toolkits.axes_grid1 import ImageGrid

# Deep Learning
import tensorflow as tf
import keras
from tensorflow.keras.preprocessing.image import ImageDataGenerator
from tensorflow.keras.backend import clear_session
from keras import backend as K
from keras.models import Sequential
from keras.layers import Dense, Dropout, Flatten
from keras.layers import Conv2D, MaxPool2D
from keras.optimizers import Adam 
from keras.callbacks import ModelCheckpoint, Callback, ReduceLROnPlateau

# Evaluate metrics
import sklearn
from sklearn.utils import shuffle
from sklearn.model_selection import KFold
from sklearn.model_selection import train_test_split

# Disable warnings
import sys
import warnings
import matplotlib.cbook
if not sys.warnoptions:
    warnings.simplefilter("ignore")
warnings.simplefilter(action='ignore', category=FutureWarning)
warnings.filterwarnings("ignore", category=FutureWarning)
warnings.filterwarnings("ignore", category=matplotlib.cbook.mplDeprecation)
%matplotlib inline

Using TensorFlow backend.


# Package versions
# Using Python 3.7.6
%reload_ext watermark
%watermark --iversions

pandas     1.0.3
PIL.Image  7.0.0
cv2.cv2    4.2.0
seaborn    0.10.0
sklearn    0.22.2
keras      2.3.1
matplotlib 3.2.1
csv        1.0
numpy      1.18.2
tensorflow 2.1.0
Data Science Academy


# Lista todos os dispositivos disponiveis
from tensorflow.python.client import device_lib
print(device_lib.list_local_devices())

[name: "/device:CPU:0"
device_type: "CPU"
memory_limit: 268435456
locality {
}
incarnation: 8763216327050890134
, name: "/device:XLA_CPU:0"
device_type: "XLA_CPU"
memory_limit: 17179869184
locality {
}
incarnation: 1473671902246478259
physical_device_desc: "device: XLA_CPU device"
, name: "/device:XLA_GPU:0"
device_type: "XLA_GPU"
memory_limit: 17179869184
locality {
}
incarnation: 16491048179918758946
physical_device_desc: "device: XLA_GPU device"
, name: "/device:XLA_GPU:1"
device_type: "XLA_GPU"
memory_limit: 17179869184
locality {
}
incarnation: 4122107778021623223
physical_device_desc: "device: XLA_GPU device"
, name: "/device:XLA_GPU:2"
device_type: "XLA_GPU"
memory_limit: 17179869184
locality {
}
incarnation: 28995238231739062
physical_device_desc: "device: XLA_GPU device"
, name: "/device:GPU:0"
device_type: "GPU"
memory_limit: 23824914842
locality {
  bus_id: 1
  links {
  }
}
incarnation: 15714529720384447878
physical_device_desc: "device: 0, name: TITAN RTX, pci bus id: 0000:0b:00.0, compute capability: 7.5"
, name: "/device:GPU:1"
device_type: "GPU"
memory_limit: 11810943796
locality {
  bus_id: 1
  links {
    link {
      device_id: 2
      type: "StreamExecutor"
      strength: 1
    }
  }
}
incarnation: 2337761275895557007
physical_device_desc: "device: 1, name: TITAN X (Pascal), pci bus id: 0000:05:00.0, compute capability: 6.1"
, name: "/device:GPU:2"
device_type: "GPU"
memory_limit: 10910800282
locality {
  bus_id: 1
  links {
    link {
      device_id: 1
      type: "StreamExecutor"
      strength: 1
    }
  }
}
incarnation: 12991201856458469678
physical_device_desc: "device: 2, name: GeForce GTX 1080 Ti, pci bus id: 0000:09:00.0, compute capability: 6.1"
]


import tensorflow as tf
print("Número Disponível de GPUs: ", len(tf.config.experimental.list_physical_devices('GPU')))

Número Disponível de GPUs:  3


# List the code for each GPU
tf.config.list_physical_devices('GPU')

[PhysicalDevice(name='/physical_device:GPU:0', device_type='GPU'),
 PhysicalDevice(name='/physical_device:GPU:1', device_type='GPU'),
 PhysicalDevice(name='/physical_device:GPU:2', device_type='GPU')]


# Directory with images
imagens_treino = '/media/datasets/DeepLearningI/Cap08/CNN02/train-jpg'
imagens_teste = '/media/datasets/DeepLearningI/Cap08/CNN02/test-jpg'


# Directory with labels
etiquetas_treino = '/media/datasets/DeepLearningI/Cap08/CNN02/labels/train_v2.csv'


# Output directory
classificacao_etiquetas_teste = 'arquivos/resultado.csv'


# Load csvs to DataFrames
dados_treino = pd.read_csv(etiquetas_treino)
dados_teste = pd.read_csv(classificacao_etiquetas_teste)


# Inspect DataFrames
dados_treino.head()


# Shape
dados_treino.shape

(40479, 2)


# Inspect DataFrames
dados_teste.head()


# Shape
dados_teste.shape

(61191, 2)


# Add the .jpg extension to image names
dados_treino['image_name'] = dados_treino['image_name'].astype(str) + '.jpg'
dados_teste['image_name'] = dados_teste['image_name'].astype(str) + '.jpg'


# Convert labels from string to list of strings
dados_teste['tags'] = dados_teste['tags'].apply(lambda x: x.split(' '))


# View DataFrame
print("\nImagens de Treino:\n")
dados_treino.head()

Imagens de Treino:


# View DataFrame
print("\nImagens de Teste (começo do arquivo):\n")
dados_teste.head()

Imagens de Teste (começo do arquivo):


# View DataFrame
print("\nImagens de Teste (final do arquivo):\n")
dados_teste.tail()

Imagens de Teste (final do arquivo):


# Shrink the test dataset size
dados_teste_final = dados_teste.iloc[:40479,]


# Store the remaining test files apart
dados_teste_restante = dados_teste.iloc[40479:,]


# Shape
dados_teste_final.shape

(40479, 2)


# Shape
dados_teste_restante.shape

(20712, 2)


# Delete the original dataset to reduce RAM usage
del dados_teste


# Train-test split and convert to np.array
X_imagens_treino = np.array(dados_treino['image_name'].tolist())
X_imagens_treino.reshape((X_imagens_treino.shape[0], 1))
y_labels_treino = np.array(dados_treino['tags'].tolist())


# X
X_imagens_treino

array(['train_0.jpg', 'train_1.jpg', 'train_2.jpg', ...,
       'train_40476.jpg', 'train_40477.jpg', 'train_40478.jpg'],
      dtype='<U15')


# Shape
X_imagens_treino.shape

(40479,)


# y
y_labels_treino

array(['haze primary', 'agriculture clear primary water', 'clear primary',
       ..., 'agriculture clear primary', 'agriculture clear primary road',
       'agriculture cultivation partly_cloudy primary'], dtype='<U92')


# Shape
y_labels_treino.shape

(40479,)


# There are 17 labels and each image has multiple labels

# Labels list
labels = []

# Loop through the labels and get all unique labels
for tag in dados_treino['tags'].values:
    labels_in_tag = tag.split(' ')
    for label in labels_in_tag:
        if label not in labels:
            labels.append(label)


# Sort the labels
labels.sort()

# Print
print('\nClasses Únicas:', len(labels))
print('\nNomes das Classes:', labels)

Classes Únicas: 17

Nomes das Classes: ['agriculture', 'artisinal_mine', 'bare_ground', 'blooming', 'blow_down', 'clear', 'cloudy', 'conventional_mine', 'cultivation', 'habitation', 'haze', 'partly_cloudy', 'primary', 'road', 'selective_logging', 'slash_burn', 'water']


# Loop through the images to check the quality of each resolution

# Figure size
plt.figure(figsize = (20, 20))

# List of resolutions
res = [32, 64, 128, 256]

# Number of images
num_imagens = 5

# All images have the same size. Resizing with OpenCV to evaluate resolutions.
for i in range(len(res)):
    for j in range(num_imagens):
        img = cv2.imread(os.path.join(imagens_treino, dados_treino['image_name'][j + 1]))
        img = cv2.resize(img, (res[i], res[i]))
        plt.subplot(len(res), num_imagens, i * num_imagens + j + 1)
        plt.imshow(cv2.cvtColor(img, cv2.COLOR_BGR2RGB))
        plt.title(dados_treino['tags'][j + 1] + "\n" + str(res[i]) + "x" + str(res[i]))
        plt.axis('off')
    
plt.show()


# Loop through train labels

# Dict to count label instances
labels_count = {}

# Loop
for tag in dados_treino['tags'].values:
    labels_in_tag = tag.split(' ')
    for label in labels_in_tag:
        if label in labels_count:
            labels_count[label] += 1
        else:
            labels_count[label] = 0


# Extract min and max
min_label = min(labels_count, key = labels_count.get)
max_label = max(labels_count, key = labels_count.get)


# Print
print(min_label + " é a classe com menor número de imagens: " + str(labels_count[min_label]))
print(max_label + " é a classe com maior número de imagens: " + str(labels_count[max_label]))

conventional_mine é a classe com menor número de imagens: 99
primary é a classe com maior número de imagens: 37512


# Figure size
plt.figure(figsize = (18, 8))

# Plot
plt.bar(range(len(labels_count)), list(labels_count.values()), align = 'center', color = 'green')
plt.xticks(range(len(labels_count)), list(labels_count.keys()), rotation = 30)
plt.show()


# Function to extract the model's learning curve
def curva_aprendizado(modelo, key = 'accuracy', ylim = (0.8, 1.01)):
    plt.figure(figsize = (15,6))
    plt.plot(modelo.history[key])
    plt.plot(modelo.history['val_' + key])
    plt.title('Curva de Aprendizado do Modelo')
    plt.ylabel(key.title())
    plt.xlabel('Epoch')
    plt.ylim(ylim)
    plt.legend(['Treino', 'Teste'], loc = 'best')
    plt.show()


# F-Beta Score function
def fbeta_score(y_true, y_pred):
    beta_squared = 4
    tp = K.sum(y_true * y_pred) + K.epsilon()
    fp = K.sum(y_pred) - tp
    fn = K.sum(y_true) - tp
    precision = tp / (tp + fp)
    recall = tp / (tp + fn)
    result = (beta_squared + 1) * (precision * recall) / (beta_squared * precision + recall + K.epsilon())
    return result


# Model hyperparameters

# Image dimensions
INPUT_SHAPE = (128, 128, 3)

# Batch Size
BATCH_SIZE = 128

# Dropout
DROPOUT_RATE = 0.1

# Number of epochs
EPOCHS = 25

# Learning rate
LR = 0.0001

# Decay rate
DECAY = 0.0001

# Number of folds for cross validation
NFOLDS = 5

# Number of each fold during training
# We start with 1 because the variable will be incremented inside the training loop
NUM_FOLD = 1

# Image classification threshold
# The sigmoid function delivers probabilities and we set the limit for one class or another
THRES = [0.2] * 17

# Directory to save models
PATH_MODELOS = 'modelos/'

# Directory to save files
PATH_ARQUIVOS = 'arquivos/'


# CNN model architecture
def cria_modelo():
    
    model = Sequential([
        
        # First convolutional layer, with two sequential convolutions
        Conv2D(32, kernel_size = (3, 3), activation = 'relu', padding = 'same', input_shape = INPUT_SHAPE),
        Conv2D(32, kernel_size = (3, 3), activation = 'relu'),
        MaxPool2D(pool_size = (2, 2)),
        Dropout(DROPOUT_RATE),

        # Second convolutional layer, with two sequential convolutions
        Conv2D(64, kernel_size = (3, 3), activation = 'relu', padding = 'same'),
        Conv2D(64, kernel_size = (3, 3), activation = 'relu'),
        MaxPool2D(pool_size = (2, 2)),
        Dropout(DROPOUT_RATE),

        # Third convolutional layer, with two sequential convolutions
        Conv2D(128, kernel_size = (3, 3), activation = 'relu', padding = 'same'),
        Conv2D(128, kernel_size = (3, 3), activation = 'relu'),
        Conv2D(128, kernel_size = (3, 3), activation = 'relu'),
        MaxPool2D(pool_size = (2, 2)),
        Dropout(DROPOUT_RATE),

        # Fourth convolutional layer, with two sequential convolutions
        Conv2D(256, kernel_size = (3, 3), activation = 'relu', padding = 'same'),
        Conv2D(256, kernel_size = (3, 3), activation = 'relu'),
        Conv2D(256, kernel_size = (3, 3), activation = 'relu'),
        MaxPool2D(pool_size = (2, 2)),
        Dropout(DROPOUT_RATE),

        Flatten(),

        Dense(17, activation = 'sigmoid') 
    ])

    # Adam optimizer
    optimizer = Adam(LR, decay = DECAY)

    return model


# Create model
modelo = cria_modelo()


# Clears the session and prints the model summary
clear_session()
modelo.summary()

Model: "sequential_1"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d_1 (Conv2D)            (None, 128, 128, 32)      896       
_________________________________________________________________
conv2d_2 (Conv2D)            (None, 126, 126, 32)      9248      
_________________________________________________________________
max_pooling2d_1 (MaxPooling2 (None, 63, 63, 32)        0         
_________________________________________________________________
dropout_1 (Dropout)          (None, 63, 63, 32)        0         
_________________________________________________________________
conv2d_3 (Conv2D)            (None, 63, 63, 64)        18496     
_________________________________________________________________
conv2d_4 (Conv2D)            (None, 61, 61, 64)        36928     
_________________________________________________________________
max_pooling2d_2 (MaxPooling2 (None, 30, 30, 64)        0         
_________________________________________________________________
dropout_2 (Dropout)          (None, 30, 30, 64)        0         
_________________________________________________________________
conv2d_5 (Conv2D)            (None, 30, 30, 128)       73856     
_________________________________________________________________
conv2d_6 (Conv2D)            (None, 28, 28, 128)       147584    
_________________________________________________________________
conv2d_7 (Conv2D)            (None, 26, 26, 128)       147584    
_________________________________________________________________
max_pooling2d_3 (MaxPooling2 (None, 13, 13, 128)       0         
_________________________________________________________________
dropout_3 (Dropout)          (None, 13, 13, 128)       0         
_________________________________________________________________
conv2d_8 (Conv2D)            (None, 13, 13, 256)       295168    
_________________________________________________________________
conv2d_9 (Conv2D)            (None, 11, 11, 256)       590080    
_________________________________________________________________
conv2d_10 (Conv2D)           (None, 9, 9, 256)         590080    
_________________________________________________________________
max_pooling2d_4 (MaxPooling2 (None, 4, 4, 256)         0         
_________________________________________________________________
dropout_4 (Dropout)          (None, 4, 4, 256)         0         
_________________________________________________________________
flatten_1 (Flatten)          (None, 4096)              0         
_________________________________________________________________
dense_1 (Dense)              (None, 17)                69649     
=================================================================
Total params: 1,979,569
Trainable params: 1,979,569
Non-trainable params: 0
_________________________________________________________________


# Empty list for the test predictions
y_teste = []


# Number of folds
folds = KFold(n_splits = NFOLDS, shuffle = True, random_state = 1).split(X_imagens_treino, y_labels_treino)


%%time

# Training loop with cross-validation
for train_index, val_index in folds:
    
    print("\nIniciando o Treinamento do Fold", NUM_FOLD)
    
    # Prepare train data
    X_train_files_fold = X_imagens_treino[train_index]
    y_train_fold = y_labels_treino[train_index]
    
    # Prepare validation data
    X_val_files_fold = X_imagens_treino[val_index]
    y_val_fold = y_labels_treino[val_index]
    
    # Convert to DataFrame
    train_df = pd.DataFrame(list(zip(X_train_files_fold, y_train_fold)), columns = ['image_name', 'tags'])
    val_df = pd.DataFrame(list(zip(X_val_files_fold, y_val_fold)), columns = ['image_name', 'tags'])
    
    # Get the labels
    train_df['tags'] = train_df['tags'].apply(lambda x: x.split(' '))
    val_df['tags'] = val_df['tags'].apply(lambda x: x.split(' '))

    print("\nDimensões dos Dados de Treino:", len(train_df))
    
    # Adjust train images
    # Based on: https://keras.io/preprocessing/image/
    train_datagen = ImageDataGenerator(rescale = 1./255, 
                                       width_shift_range = 0.2,
                                       height_shift_range = 0.2,
                                       horizontal_flip = True,
                                       vertical_flip = True)
    
    # Adjust test images
    train_generator = train_datagen.flow_from_dataframe(train_df,
                                                        directory = imagens_treino,
                                                        x_col = 'image_name',
                                                        y_col = 'tags',
                                                        target_size = (INPUT_SHAPE[0], INPUT_SHAPE[1]), # 128 x 128
                                                        class_mode = 'categorical',
                                                        batch_size = BATCH_SIZE,
                                                        classes = labels,)
    
    print("\nDimensões dos Dados de Validação:", len(val_df))
    
    # Adjust valid images
    val_datagen = ImageDataGenerator(rescale = 1./255)
    
    # Prepare validation dataset
    val_generator = val_datagen.flow_from_dataframe(val_df,
                                                    directory = imagens_treino,
                                                    x_col = 'image_name',
                                                    y_col = 'tags',
                                                    target_size = (INPUT_SHAPE[0], INPUT_SHAPE[1]), # 128 x 128
                                                    class_mode = 'categorical',
                                                    batch_size = BATCH_SIZE,
                                                    classes = labels,)
    
    print("\nDimensões dos Dados de Teste:", len(dados_teste_final))
    
    # Adjust test images
    test_datagen = ImageDataGenerator(rescale = 1./255)
    
    # Prepare test dataset
    test_generator = test_datagen.flow_from_dataframe(dados_teste_final,
                                                      directory = imagens_teste,
                                                      x_col = 'image_name',
                                                      y_col = 'tags',
                                                      target_size = (INPUT_SHAPE[0], INPUT_SHAPE[1]), # 128 x 128
                                                      class_mode = 'categorical',
                                                      batch_size = BATCH_SIZE,
                                                      classes = labels,
                                                      shuffle = False,)
    
    # Clear session (free GPU memory)
    clear_session()
    
    # Create model
    modelo_cnn_satelite = cria_modelo()
    
    # Learning Rate and Adam Optinmizer
    adam = Adam(learning_rate = LR)
    
    # Loss functions
    func_loss = 'binary_crossentropy'
    
    # Compile model
    modelo_cnn_satelite.compile(optimizer = adam, loss = func_loss, metrics = [fbeta_score, 'accuracy'])
    
    # Path to save the model after each fold
    modelo_fold_salvo = os.path.join(PATH_MODELOS, 'pesos_fold_' + str(NUM_FOLD) + '.h5')
    
    # Callbacks to checkpoint the model and reduce learning rate on plateau
    # https://www.tensorflow.org/api_docs/python/tf/keras/callbacks/ReduceLROnPlateau
    callbacks = [ModelCheckpoint(modelo_fold_salvo, 
                                 monitor = 'fbeta_score', 
                                 save_best_only = True, 
                                 mode = 'max'),
                 ReduceLROnPlateau(monitor = 'loss', 
                                   factor = 0.1, 
                                   patience = 3, 
                                   mode = 'min', 
                                   min_lr = 0.000001,
                                   verbose = 1)]
    
    # Trein model
    modelo_cnn_satelite.fit_generator(train_generator, 
                                      epochs = EPOCHS, 
                                      validation_data = val_generator, 
                                      callbacks = callbacks)

    # Load model from disk
    modelo_cnn_satelite.load_weights(modelo_fold_salvo)

    # Predict
    print("\nModelo fazendo a classificação nos dados de teste. Seja paciente e aguarde...")
    previsao_teste = modelo_cnn_satelite.predict_generator(test_generator)
    
    # Store results
    y_teste.append(previsao_teste)
    
    # Go to next cross-validation fold
    print("\nTérmino do Fold ", NUM_FOLD, " da Validação Cruzada.")
    NUM_FOLD += 1

print("\nTreinamento Concluído.")

Iniciando o Treinamento do Fold 1

Dimensões dos Dados de Treino: 32383
Found 32383 validated image filenames belonging to 17 classes.

Dimensões dos Dados de Validação: 8096
Found 8096 validated image filenames belonging to 17 classes.

Dimensões dos Dados de Teste: 40479
Found 40479 validated image filenames belonging to 17 classes.
Epoch 1/25
253/253 [==============================] - 152s 602ms/step - loss: 0.2766 - fbeta_score: 0.5452 - accuracy: 0.9030 - val_loss: 0.2300 - val_fbeta_score: 0.5863 - val_accuracy: 0.9122
Epoch 2/25
253/253 [==============================] - 146s 576ms/step - loss: 0.2177 - fbeta_score: 0.6129 - accuracy: 0.9143 - val_loss: 0.1973 - val_fbeta_score: 0.6298 - val_accuracy: 0.9183
Epoch 3/25
253/253 [==============================] - 147s 581ms/step - loss: 0.2022 - fbeta_score: 0.6426 - accuracy: 0.9200 - val_loss: 0.1943 - val_fbeta_score: 0.6615 - val_accuracy: 0.9249
Epoch 4/25
253/253 [==============================] - 145s 573ms/step - loss: 0.1890 - fbeta_score: 0.6686 - accuracy: 0.9254 - val_loss: 0.2025 - val_fbeta_score: 0.6881 - val_accuracy: 0.9277
Epoch 5/25
253/253 [==============================] - 147s 583ms/step - loss: 0.1767 - fbeta_score: 0.6930 - accuracy: 0.9307 - val_loss: 0.2211 - val_fbeta_score: 0.7129 - val_accuracy: 0.9354
Epoch 6/25
253/253 [==============================] - 146s 576ms/step - loss: 0.1680 - fbeta_score: 0.7102 - accuracy: 0.9344 - val_loss: 0.2049 - val_fbeta_score: 0.7093 - val_accuracy: 0.9363
Epoch 7/25
253/253 [==============================] - 146s 579ms/step - loss: 0.1618 - fbeta_score: 0.7213 - accuracy: 0.9370 - val_loss: 0.1628 - val_fbeta_score: 0.7267 - val_accuracy: 0.9390
Epoch 8/25
253/253 [==============================] - 145s 575ms/step - loss: 0.1586 - fbeta_score: 0.7272 - accuracy: 0.9382 - val_loss: 0.1622 - val_fbeta_score: 0.7401 - val_accuracy: 0.9420
Epoch 9/25
253/253 [==============================] - 146s 577ms/step - loss: 0.1555 - fbeta_score: 0.7326 - accuracy: 0.9395 - val_loss: 0.1630 - val_fbeta_score: 0.7333 - val_accuracy: 0.9407
Epoch 10/25
253/253 [==============================] - 146s 576ms/step - loss: 0.1523 - fbeta_score: 0.7385 - accuracy: 0.9407 - val_loss: 0.1389 - val_fbeta_score: 0.7365 - val_accuracy: 0.9417
Epoch 11/25
253/253 [==============================] - 145s 575ms/step - loss: 0.1502 - fbeta_score: 0.7413 - accuracy: 0.9411 - val_loss: 0.1641 - val_fbeta_score: 0.7595 - val_accuracy: 0.9426
Epoch 12/25
253/253 [==============================] - 146s 576ms/step - loss: 0.1490 - fbeta_score: 0.7440 - accuracy: 0.9418 - val_loss: 0.1328 - val_fbeta_score: 0.7510 - val_accuracy: 0.9435
Epoch 13/25
253/253 [==============================] - 146s 576ms/step - loss: 0.1467 - fbeta_score: 0.7476 - accuracy: 0.9425 - val_loss: 0.1447 - val_fbeta_score: 0.7510 - val_accuracy: 0.9428
Epoch 14/25
253/253 [==============================] - 145s 573ms/step - loss: 0.1449 - fbeta_score: 0.7501 - accuracy: 0.9431 - val_loss: 0.1546 - val_fbeta_score: 0.7704 - val_accuracy: 0.9453
Epoch 15/25
253/253 [==============================] - 145s 573ms/step - loss: 0.1434 - fbeta_score: 0.7535 - accuracy: 0.9439 - val_loss: 0.1645 - val_fbeta_score: 0.7705 - val_accuracy: 0.9446
Epoch 16/25
253/253 [==============================] - 146s 576ms/step - loss: 0.1417 - fbeta_score: 0.7560 - accuracy: 0.9440 - val_loss: 0.1345 - val_fbeta_score: 0.7612 - val_accuracy: 0.9445
Epoch 17/25
253/253 [==============================] - 145s 574ms/step - loss: 0.1404 - fbeta_score: 0.7581 - accuracy: 0.9449 - val_loss: 0.1495 - val_fbeta_score: 0.7692 - val_accuracy: 0.9458
Epoch 18/25
253/253 [==============================] - 146s 578ms/step - loss: 0.1390 - fbeta_score: 0.7608 - accuracy: 0.9454 - val_loss: 0.1086 - val_fbeta_score: 0.7810 - val_accuracy: 0.9467
Epoch 19/25
253/253 [==============================] - 146s 576ms/step - loss: 0.1374 - fbeta_score: 0.7635 - accuracy: 0.9460 - val_loss: 0.1208 - val_fbeta_score: 0.7765 - val_accuracy: 0.9487
Epoch 20/25
253/253 [==============================] - 145s 571ms/step - loss: 0.1374 - fbeta_score: 0.7635 - accuracy: 0.9460 - val_loss: 0.1183 - val_fbeta_score: 0.7837 - val_accuracy: 0.9485
Epoch 21/25
253/253 [==============================] - 146s 577ms/step - loss: 0.1367 - fbeta_score: 0.7649 - accuracy: 0.9463 - val_loss: 0.1377 - val_fbeta_score: 0.7619 - val_accuracy: 0.9475
Epoch 22/25
253/253 [==============================] - 145s 574ms/step - loss: 0.1349 - fbeta_score: 0.7680 - accuracy: 0.9471 - val_loss: 0.1591 - val_fbeta_score: 0.7751 - val_accuracy: 0.9473
Epoch 23/25
253/253 [==============================] - 147s 579ms/step - loss: 0.1336 - fbeta_score: 0.7707 - accuracy: 0.9478 - val_loss: 0.0882 - val_fbeta_score: 0.7795 - val_accuracy: 0.9491
Epoch 24/25
253/253 [==============================] - 146s 578ms/step - loss: 0.1334 - fbeta_score: 0.7706 - accuracy: 0.9479 - val_loss: 0.1252 - val_fbeta_score: 0.7851 - val_accuracy: 0.9484
Epoch 25/25
253/253 [==============================] - 146s 579ms/step - loss: 0.1322 - fbeta_score: 0.7731 - accuracy: 0.9481 - val_loss: 0.1083 - val_fbeta_score: 0.7899 - val_accuracy: 0.9494

Modelo fazendo a classificação nos dados de teste. Seja paciente e aguarde...

Término do Fold  1  da Validação Cruzada.

Iniciando o Treinamento do Fold 2

Dimensões dos Dados de Treino: 32383
Found 32383 validated image filenames belonging to 17 classes.

Dimensões dos Dados de Validação: 8096
Found 8096 validated image filenames belonging to 17 classes.

Dimensões dos Dados de Teste: 40479
Found 40479 validated image filenames belonging to 17 classes.
Epoch 1/25
253/253 [==============================] - 152s 601ms/step - loss: 0.2777 - fbeta_score: 0.5462 - accuracy: 0.9020 - val_loss: 0.2567 - val_fbeta_score: 0.5752 - val_accuracy: 0.9102
Epoch 2/25
253/253 [==============================] - 145s 575ms/step - loss: 0.2220 - fbeta_score: 0.6041 - accuracy: 0.9118 - val_loss: 0.2268 - val_fbeta_score: 0.6008 - val_accuracy: 0.9129
Epoch 3/25
253/253 [==============================] - 145s 575ms/step - loss: 0.2075 - fbeta_score: 0.6335 - accuracy: 0.9185 - val_loss: 0.2014 - val_fbeta_score: 0.6205 - val_accuracy: 0.9181
Epoch 4/25
253/253 [==============================] - 146s 578ms/step - loss: 0.1965 - fbeta_score: 0.6531 - accuracy: 0.9221 - val_loss: 0.2377 - val_fbeta_score: 0.6625 - val_accuracy: 0.9262
Epoch 5/25
253/253 [==============================] - 146s 577ms/step - loss: 0.1843 - fbeta_score: 0.6776 - accuracy: 0.9273 - val_loss: 0.1797 - val_fbeta_score: 0.6925 - val_accuracy: 0.9309
Epoch 6/25
253/253 [==============================] - 146s 578ms/step - loss: 0.1722 - fbeta_score: 0.7011 - accuracy: 0.9325 - val_loss: 0.1445 - val_fbeta_score: 0.6981 - val_accuracy: 0.9327
Epoch 7/25
253/253 [==============================] - 146s 578ms/step - loss: 0.1648 - fbeta_score: 0.7154 - accuracy: 0.9358 - val_loss: 0.1755 - val_fbeta_score: 0.7327 - val_accuracy: 0.9343
Epoch 8/25
253/253 [==============================] - 146s 579ms/step - loss: 0.1594 - fbeta_score: 0.7259 - accuracy: 0.9381 - val_loss: 0.1627 - val_fbeta_score: 0.7353 - val_accuracy: 0.9399
Epoch 9/25
253/253 [==============================] - 148s 583ms/step - loss: 0.1563 - fbeta_score: 0.7318 - accuracy: 0.9393 - val_loss: 0.1186 - val_fbeta_score: 0.7451 - val_accuracy: 0.9398
Epoch 10/25
253/253 [==============================] - 146s 578ms/step - loss: 0.1522 - fbeta_score: 0.7387 - accuracy: 0.9409 - val_loss: 0.1003 - val_fbeta_score: 0.7557 - val_accuracy: 0.9411
Epoch 11/25
253/253 [==============================] - 145s 575ms/step - loss: 0.1493 - fbeta_score: 0.7434 - accuracy: 0.9417 - val_loss: 0.1512 - val_fbeta_score: 0.7514 - val_accuracy: 0.9434
Epoch 12/25
253/253 [==============================] - 147s 579ms/step - loss: 0.1477 - fbeta_score: 0.7467 - accuracy: 0.9421 - val_loss: 0.1477 - val_fbeta_score: 0.7567 - val_accuracy: 0.9436
Epoch 13/25
253/253 [==============================] - 146s 577ms/step - loss: 0.1461 - fbeta_score: 0.7492 - accuracy: 0.9428 - val_loss: 0.1895 - val_fbeta_score: 0.7670 - val_accuracy: 0.9407
Epoch 14/25
253/253 [==============================] - 147s 580ms/step - loss: 0.1442 - fbeta_score: 0.7527 - accuracy: 0.9435 - val_loss: 0.1587 - val_fbeta_score: 0.7735 - val_accuracy: 0.9445
Epoch 15/25
253/253 [==============================] - 146s 575ms/step - loss: 0.1432 - fbeta_score: 0.7541 - accuracy: 0.9440 - val_loss: 0.1339 - val_fbeta_score: 0.7739 - val_accuracy: 0.9451
Epoch 16/25
253/253 [==============================] - 146s 576ms/step - loss: 0.1418 - fbeta_score: 0.7569 - accuracy: 0.9446 - val_loss: 0.1293 - val_fbeta_score: 0.7700 - val_accuracy: 0.9462
Epoch 17/25
253/253 [==============================] - 147s 581ms/step - loss: 0.1405 - fbeta_score: 0.7587 - accuracy: 0.9450 - val_loss: 0.0942 - val_fbeta_score: 0.7727 - val_accuracy: 0.9436
Epoch 18/25
253/253 [==============================] - 145s 573ms/step - loss: 0.1388 - fbeta_score: 0.7617 - accuracy: 0.9458 - val_loss: 0.0982 - val_fbeta_score: 0.7714 - val_accuracy: 0.9485
Epoch 19/25
253/253 [==============================] - 145s 574ms/step - loss: 0.1367 - fbeta_score: 0.7656 - accuracy: 0.9466 - val_loss: 0.1025 - val_fbeta_score: 0.7776 - val_accuracy: 0.9474
Epoch 20/25
253/253 [==============================] - 145s 574ms/step - loss: 0.1362 - fbeta_score: 0.7667 - accuracy: 0.9470 - val_loss: 0.1500 - val_fbeta_score: 0.7840 - val_accuracy: 0.9475
Epoch 21/25
253/253 [==============================] - 145s 572ms/step - loss: 0.1354 - fbeta_score: 0.7680 - accuracy: 0.9472 - val_loss: 0.1277 - val_fbeta_score: 0.7810 - val_accuracy: 0.9491
Epoch 22/25
253/253 [==============================] - 146s 577ms/step - loss: 0.1344 - fbeta_score: 0.7701 - accuracy: 0.9479 - val_loss: 0.1337 - val_fbeta_score: 0.7839 - val_accuracy: 0.9504
Epoch 23/25
253/253 [==============================] - 146s 577ms/step - loss: 0.1329 - fbeta_score: 0.7725 - accuracy: 0.9481 - val_loss: 0.1010 - val_fbeta_score: 0.7800 - val_accuracy: 0.9486
Epoch 24/25
253/253 [==============================] - 146s 578ms/step - loss: 0.1323 - fbeta_score: 0.7738 - accuracy: 0.9486 - val_loss: 0.1064 - val_fbeta_score: 0.7843 - val_accuracy: 0.9484
Epoch 25/25
253/253 [==============================] - 146s 576ms/step - loss: 0.1316 - fbeta_score: 0.7746 - accuracy: 0.9489 - val_loss: 0.1202 - val_fbeta_score: 0.7870 - val_accuracy: 0.9496

Modelo fazendo a classificação nos dados de teste. Seja paciente e aguarde...

Término do Fold  2  da Validação Cruzada.

Iniciando o Treinamento do Fold 3

Dimensões dos Dados de Treino: 32383
Found 32383 validated image filenames belonging to 17 classes.

Dimensões dos Dados de Validação: 8096
Found 8096 validated image filenames belonging to 17 classes.

Dimensões dos Dados de Teste: 40479
Found 40479 validated image filenames belonging to 17 classes.
Epoch 1/25
253/253 [==============================] - 152s 600ms/step - loss: 0.2708 - fbeta_score: 0.5518 - accuracy: 0.9043 - val_loss: 0.2716 - val_fbeta_score: 0.5613 - val_accuracy: 0.9045
Epoch 2/25
253/253 [==============================] - 147s 580ms/step - loss: 0.2185 - fbeta_score: 0.6116 - accuracy: 0.9137 - val_loss: 0.2327 - val_fbeta_score: 0.5864 - val_accuracy: 0.9142
Epoch 3/25
253/253 [==============================] - 146s 578ms/step - loss: 0.2057 - fbeta_score: 0.6366 - accuracy: 0.9192 - val_loss: 0.2413 - val_fbeta_score: 0.6509 - val_accuracy: 0.9198
Epoch 4/25
253/253 [==============================] - 146s 576ms/step - loss: 0.1929 - fbeta_score: 0.6615 - accuracy: 0.9242 - val_loss: 0.1781 - val_fbeta_score: 0.6756 - val_accuracy: 0.9273
Epoch 5/25
253/253 [==============================] - 146s 578ms/step - loss: 0.1809 - fbeta_score: 0.6849 - accuracy: 0.9292 - val_loss: 0.2149 - val_fbeta_score: 0.6999 - val_accuracy: 0.9313
Epoch 6/25
253/253 [==============================] - 147s 579ms/step - loss: 0.1724 - fbeta_score: 0.7016 - accuracy: 0.9326 - val_loss: 0.1883 - val_fbeta_score: 0.7129 - val_accuracy: 0.9333
Epoch 7/25
253/253 [==============================] - 146s 577ms/step - loss: 0.1669 - fbeta_score: 0.7123 - accuracy: 0.9351 - val_loss: 0.1389 - val_fbeta_score: 0.7213 - val_accuracy: 0.9356
Epoch 8/25
253/253 [==============================] - 146s 577ms/step - loss: 0.1614 - fbeta_score: 0.7221 - accuracy: 0.9371 - val_loss: 0.1585 - val_fbeta_score: 0.7201 - val_accuracy: 0.9386
Epoch 9/25
253/253 [==============================] - 145s 572ms/step - loss: 0.1587 - fbeta_score: 0.7269 - accuracy: 0.9381 - val_loss: 0.1479 - val_fbeta_score: 0.7243 - val_accuracy: 0.9392
Epoch 10/25
253/253 [==============================] - 147s 582ms/step - loss: 0.1553 - fbeta_score: 0.7328 - accuracy: 0.9395 - val_loss: 0.1281 - val_fbeta_score: 0.7506 - val_accuracy: 0.9401
Epoch 11/25
253/253 [==============================] - 145s 575ms/step - loss: 0.1532 - fbeta_score: 0.7370 - accuracy: 0.9401 - val_loss: 0.1308 - val_fbeta_score: 0.7413 - val_accuracy: 0.9424
Epoch 12/25
253/253 [==============================] - 145s 575ms/step - loss: 0.1506 - fbeta_score: 0.7411 - accuracy: 0.9409 - val_loss: 0.1171 - val_fbeta_score: 0.7444 - val_accuracy: 0.9427
Epoch 13/25
253/253 [==============================] - 146s 576ms/step - loss: 0.1494 - fbeta_score: 0.7426 - accuracy: 0.9415 - val_loss: 0.1106 - val_fbeta_score: 0.7635 - val_accuracy: 0.9430
Epoch 14/25
253/253 [==============================] - 145s 573ms/step - loss: 0.1469 - fbeta_score: 0.7469 - accuracy: 0.9424 - val_loss: 0.1599 - val_fbeta_score: 0.7603 - val_accuracy: 0.9450
Epoch 15/25
253/253 [==============================] - 146s 579ms/step - loss: 0.1451 - fbeta_score: 0.7505 - accuracy: 0.9432 - val_loss: 0.2192 - val_fbeta_score: 0.7515 - val_accuracy: 0.9462
Epoch 16/25
253/253 [==============================] - 145s 573ms/step - loss: 0.1436 - fbeta_score: 0.7525 - accuracy: 0.9438 - val_loss: 0.0913 - val_fbeta_score: 0.7662 - val_accuracy: 0.9430
Epoch 17/25
253/253 [==============================] - 145s 574ms/step - loss: 0.1416 - fbeta_score: 0.7565 - accuracy: 0.9446 - val_loss: 0.1277 - val_fbeta_score: 0.7566 - val_accuracy: 0.9452
Epoch 18/25
253/253 [==============================] - 146s 579ms/step - loss: 0.1403 - fbeta_score: 0.7585 - accuracy: 0.9448 - val_loss: 0.1682 - val_fbeta_score: 0.7593 - val_accuracy: 0.9462
Epoch 19/25
253/253 [==============================] - 146s 575ms/step - loss: 0.1398 - fbeta_score: 0.7594 - accuracy: 0.9451 - val_loss: 0.1388 - val_fbeta_score: 0.7717 - val_accuracy: 0.9483
Epoch 20/25
253/253 [==============================] - 145s 574ms/step - loss: 0.1380 - fbeta_score: 0.7628 - accuracy: 0.9462 - val_loss: 0.1300 - val_fbeta_score: 0.7679 - val_accuracy: 0.9460
Epoch 21/25
253/253 [==============================] - 145s 573ms/step - loss: 0.1365 - fbeta_score: 0.7657 - accuracy: 0.9467 - val_loss: 0.1508 - val_fbeta_score: 0.7723 - val_accuracy: 0.9487
Epoch 22/25
253/253 [==============================] - 146s 579ms/step - loss: 0.1359 - fbeta_score: 0.7665 - accuracy: 0.9468 - val_loss: 0.1578 - val_fbeta_score: 0.7815 - val_accuracy: 0.9472
Epoch 23/25
253/253 [==============================] - 144s 568ms/step - loss: 0.1346 - fbeta_score: 0.7690 - accuracy: 0.9475 - val_loss: 0.1014 - val_fbeta_score: 0.7762 - val_accuracy: 0.9481
Epoch 24/25
253/253 [==============================] - 146s 575ms/step - loss: 0.1339 - fbeta_score: 0.7703 - accuracy: 0.9478 - val_loss: 0.1097 - val_fbeta_score: 0.7821 - val_accuracy: 0.9499
Epoch 25/25
253/253 [==============================] - 145s 571ms/step - loss: 0.1339 - fbeta_score: 0.7701 - accuracy: 0.9478 - val_loss: 0.1253 - val_fbeta_score: 0.7768 - val_accuracy: 0.9503

Modelo fazendo a classificação nos dados de teste. Seja paciente e aguarde...

Término do Fold  3  da Validação Cruzada.

Iniciando o Treinamento do Fold 4

Dimensões dos Dados de Treino: 32383
Found 32383 validated image filenames belonging to 17 classes.

Dimensões dos Dados de Validação: 8096
Found 8096 validated image filenames belonging to 17 classes.

Dimensões dos Dados de Teste: 40479
Found 40479 validated image filenames belonging to 17 classes.
Epoch 1/25
253/253 [==============================] - 151s 598ms/step - loss: 0.2732 - fbeta_score: 0.5512 - accuracy: 0.9047 - val_loss: 0.2206 - val_fbeta_score: 0.5700 - val_accuracy: 0.9035
Epoch 2/25
253/253 [==============================] - 146s 575ms/step - loss: 0.2198 - fbeta_score: 0.6088 - accuracy: 0.9134 - val_loss: 0.2312 - val_fbeta_score: 0.5970 - val_accuracy: 0.9137
Epoch 3/25
253/253 [==============================] - 147s 580ms/step - loss: 0.2098 - fbeta_score: 0.6290 - accuracy: 0.9177 - val_loss: 0.1696 - val_fbeta_score: 0.6268 - val_accuracy: 0.9190
Epoch 4/25
253/253 [==============================] - 145s 572ms/step - loss: 0.1985 - fbeta_score: 0.6492 - accuracy: 0.9209 - val_loss: 0.1950 - val_fbeta_score: 0.6461 - val_accuracy: 0.9218
Epoch 5/25
253/253 [==============================] - 147s 583ms/step - loss: 0.1836 - fbeta_score: 0.6782 - accuracy: 0.9274 - val_loss: 0.2005 - val_fbeta_score: 0.6788 - val_accuracy: 0.9303
Epoch 6/25
253/253 [==============================] - 147s 579ms/step - loss: 0.1736 - fbeta_score: 0.6982 - accuracy: 0.9318 - val_loss: 0.1900 - val_fbeta_score: 0.6992 - val_accuracy: 0.9353
Epoch 7/25
253/253 [==============================] - 146s 578ms/step - loss: 0.1670 - fbeta_score: 0.7110 - accuracy: 0.9346 - val_loss: 0.1471 - val_fbeta_score: 0.7283 - val_accuracy: 0.9362
Epoch 8/25
253/253 [==============================] - 146s 578ms/step - loss: 0.1614 - fbeta_score: 0.7219 - accuracy: 0.9371 - val_loss: 0.1270 - val_fbeta_score: 0.7402 - val_accuracy: 0.9379
Epoch 9/25
253/253 [==============================] - 147s 582ms/step - loss: 0.1573 - fbeta_score: 0.7290 - accuracy: 0.9387 - val_loss: 0.1837 - val_fbeta_score: 0.7290 - val_accuracy: 0.9406
Epoch 10/25
253/253 [==============================] - 146s 578ms/step - loss: 0.1528 - fbeta_score: 0.7367 - accuracy: 0.9401 - val_loss: 0.0832 - val_fbeta_score: 0.7483 - val_accuracy: 0.9405
Epoch 11/25
253/253 [==============================] - 147s 582ms/step - loss: 0.1503 - fbeta_score: 0.7413 - accuracy: 0.9411 - val_loss: 0.1397 - val_fbeta_score: 0.7419 - val_accuracy: 0.9428
Epoch 12/25
253/253 [==============================] - 148s 584ms/step - loss: 0.1478 - fbeta_score: 0.7450 - accuracy: 0.9420 - val_loss: 0.1768 - val_fbeta_score: 0.7621 - val_accuracy: 0.9405
Epoch 13/25
253/253 [==============================] - 147s 582ms/step - loss: 0.1457 - fbeta_score: 0.7489 - accuracy: 0.9427 - val_loss: 0.1257 - val_fbeta_score: 0.7530 - val_accuracy: 0.9447
Epoch 14/25
253/253 [==============================] - 147s 579ms/step - loss: 0.1442 - fbeta_score: 0.7514 - accuracy: 0.9431 - val_loss: 0.1282 - val_fbeta_score: 0.7744 - val_accuracy: 0.9433
Epoch 15/25
253/253 [==============================] - 147s 580ms/step - loss: 0.1420 - fbeta_score: 0.7553 - accuracy: 0.9444 - val_loss: 0.0902 - val_fbeta_score: 0.7634 - val_accuracy: 0.9455
Epoch 16/25
253/253 [==============================] - 147s 580ms/step - loss: 0.1417 - fbeta_score: 0.7565 - accuracy: 0.9444 - val_loss: 0.1477 - val_fbeta_score: 0.7614 - val_accuracy: 0.9465
Epoch 17/25
253/253 [==============================] - 146s 578ms/step - loss: 0.1408 - fbeta_score: 0.7577 - accuracy: 0.9448 - val_loss: 0.1759 - val_fbeta_score: 0.7640 - val_accuracy: 0.9454
Epoch 18/25
253/253 [==============================] - 146s 578ms/step - loss: 0.1385 - fbeta_score: 0.7617 - accuracy: 0.9454 - val_loss: 0.1548 - val_fbeta_score: 0.7763 - val_accuracy: 0.9468
Epoch 19/25
253/253 [==============================] - 145s 573ms/step - loss: 0.1380 - fbeta_score: 0.7632 - accuracy: 0.9461 - val_loss: 0.1193 - val_fbeta_score: 0.7690 - val_accuracy: 0.9484
Epoch 20/25
253/253 [==============================] - 147s 581ms/step - loss: 0.1362 - fbeta_score: 0.7655 - accuracy: 0.9466 - val_loss: 0.1316 - val_fbeta_score: 0.7785 - val_accuracy: 0.9458
Epoch 21/25
253/253 [==============================] - 146s 576ms/step - loss: 0.1356 - fbeta_score: 0.7673 - accuracy: 0.9471 - val_loss: 0.0925 - val_fbeta_score: 0.7770 - val_accuracy: 0.9488
Epoch 22/25
253/253 [==============================] - 145s 574ms/step - loss: 0.1335 - fbeta_score: 0.7707 - accuracy: 0.9476 - val_loss: 0.1034 - val_fbeta_score: 0.7883 - val_accuracy: 0.9496
Epoch 23/25
253/253 [==============================] - 146s 579ms/step - loss: 0.1326 - fbeta_score: 0.7724 - accuracy: 0.9483 - val_loss: 0.0823 - val_fbeta_score: 0.7807 - val_accuracy: 0.9485
Epoch 24/25
253/253 [==============================] - 144s 570ms/step - loss: 0.1317 - fbeta_score: 0.7740 - accuracy: 0.9484 - val_loss: 0.1295 - val_fbeta_score: 0.7923 - val_accuracy: 0.9508
Epoch 25/25
253/253 [==============================] - 146s 576ms/step - loss: 0.1303 - fbeta_score: 0.7771 - accuracy: 0.9492 - val_loss: 0.1078 - val_fbeta_score: 0.7826 - val_accuracy: 0.9457

Modelo fazendo a classificação nos dados de teste. Seja paciente e aguarde...

Término do Fold  4  da Validação Cruzada.

Iniciando o Treinamento do Fold 5

Dimensões dos Dados de Treino: 32384
Found 32384 validated image filenames belonging to 17 classes.

Dimensões dos Dados de Validação: 8095
Found 8095 validated image filenames belonging to 17 classes.

Dimensões dos Dados de Teste: 40479
Found 40479 validated image filenames belonging to 17 classes.
Epoch 1/25
253/253 [==============================] - 152s 599ms/step - loss: 0.2753 - fbeta_score: 0.5466 - accuracy: 0.9013 - val_loss: 0.2264 - val_fbeta_score: 0.5667 - val_accuracy: 0.9116
Epoch 2/25
253/253 [==============================] - 145s 575ms/step - loss: 0.2213 - fbeta_score: 0.6058 - accuracy: 0.9130 - val_loss: 0.2295 - val_fbeta_score: 0.6052 - val_accuracy: 0.9158
Epoch 3/25
253/253 [==============================] - 147s 583ms/step - loss: 0.2069 - fbeta_score: 0.6352 - accuracy: 0.9187 - val_loss: 0.2025 - val_fbeta_score: 0.6370 - val_accuracy: 0.9207
Epoch 4/25
253/253 [==============================] - 147s 580ms/step - loss: 0.1940 - fbeta_score: 0.6585 - accuracy: 0.9228 - val_loss: 0.1936 - val_fbeta_score: 0.6646 - val_accuracy: 0.9272
Epoch 5/25
253/253 [==============================] - 147s 579ms/step - loss: 0.1794 - fbeta_score: 0.6873 - accuracy: 0.9294 - val_loss: 0.2690 - val_fbeta_score: 0.6927 - val_accuracy: 0.9287
Epoch 6/25
253/253 [==============================] - 147s 580ms/step - loss: 0.1688 - fbeta_score: 0.7087 - accuracy: 0.9341 - val_loss: 0.1892 - val_fbeta_score: 0.7084 - val_accuracy: 0.9381
Epoch 7/25
253/253 [==============================] - 147s 579ms/step - loss: 0.1624 - fbeta_score: 0.7211 - accuracy: 0.9370 - val_loss: 0.1867 - val_fbeta_score: 0.7232 - val_accuracy: 0.9390
Epoch 8/25
253/253 [==============================] - 147s 581ms/step - loss: 0.1573 - fbeta_score: 0.7300 - accuracy: 0.9385 - val_loss: 0.1405 - val_fbeta_score: 0.7413 - val_accuracy: 0.9404
Epoch 9/25
253/253 [==============================] - 146s 577ms/step - loss: 0.1555 - fbeta_score: 0.7329 - accuracy: 0.9395 - val_loss: 0.1963 - val_fbeta_score: 0.7433 - val_accuracy: 0.9420
Epoch 10/25
253/253 [==============================] - 146s 579ms/step - loss: 0.1508 - fbeta_score: 0.7411 - accuracy: 0.9411 - val_loss: 0.1383 - val_fbeta_score: 0.7441 - val_accuracy: 0.9442
Epoch 11/25
253/253 [==============================] - 146s 578ms/step - loss: 0.1484 - fbeta_score: 0.7454 - accuracy: 0.9418 - val_loss: 0.1601 - val_fbeta_score: 0.7433 - val_accuracy: 0.9429
Epoch 12/25
253/253 [==============================] - 146s 575ms/step - loss: 0.1479 - fbeta_score: 0.7457 - accuracy: 0.9422 - val_loss: 0.1565 - val_fbeta_score: 0.7586 - val_accuracy: 0.9456
Epoch 13/25
253/253 [==============================] - 145s 574ms/step - loss: 0.1447 - fbeta_score: 0.7517 - accuracy: 0.9435 - val_loss: 0.1563 - val_fbeta_score: 0.7626 - val_accuracy: 0.9459
Epoch 14/25
253/253 [==============================] - 147s 579ms/step - loss: 0.1425 - fbeta_score: 0.7554 - accuracy: 0.9443 - val_loss: 0.1274 - val_fbeta_score: 0.7650 - val_accuracy: 0.9466
Epoch 15/25
253/253 [==============================] - 146s 577ms/step - loss: 0.1407 - fbeta_score: 0.7580 - accuracy: 0.9449 - val_loss: 0.1201 - val_fbeta_score: 0.7750 - val_accuracy: 0.9458
Epoch 16/25
253/253 [==============================] - 146s 577ms/step - loss: 0.1396 - fbeta_score: 0.7597 - accuracy: 0.9447 - val_loss: 0.1447 - val_fbeta_score: 0.7641 - val_accuracy: 0.9462
Epoch 17/25
253/253 [==============================] - 147s 581ms/step - loss: 0.1387 - fbeta_score: 0.7613 - accuracy: 0.9457 - val_loss: 0.1385 - val_fbeta_score: 0.7715 - val_accuracy: 0.9471
Epoch 18/25
253/253 [==============================] - 145s 574ms/step - loss: 0.1373 - fbeta_score: 0.7640 - accuracy: 0.9458 - val_loss: 0.1304 - val_fbeta_score: 0.7856 - val_accuracy: 0.9477
Epoch 19/25
253/253 [==============================] - 145s 574ms/step - loss: 0.1355 - fbeta_score: 0.7671 - accuracy: 0.9468 - val_loss: 0.1494 - val_fbeta_score: 0.7920 - val_accuracy: 0.9486
Epoch 20/25
253/253 [==============================] - 146s 576ms/step - loss: 0.1342 - fbeta_score: 0.7694 - accuracy: 0.9471 - val_loss: 0.1256 - val_fbeta_score: 0.7891 - val_accuracy: 0.9497
Epoch 21/25
253/253 [==============================] - 146s 577ms/step - loss: 0.1334 - fbeta_score: 0.7713 - accuracy: 0.9477 - val_loss: 0.1153 - val_fbeta_score: 0.7730 - val_accuracy: 0.9500
Epoch 22/25
253/253 [==============================] - 146s 578ms/step - loss: 0.1323 - fbeta_score: 0.7728 - accuracy: 0.9479 - val_loss: 0.1466 - val_fbeta_score: 0.7890 - val_accuracy: 0.9501
Epoch 23/25
253/253 [==============================] - 146s 576ms/step - loss: 0.1320 - fbeta_score: 0.7738 - accuracy: 0.9482 - val_loss: 0.1478 - val_fbeta_score: 0.7870 - val_accuracy: 0.9497
Epoch 24/25
253/253 [==============================] - 145s 574ms/step - loss: 0.1308 - fbeta_score: 0.7757 - accuracy: 0.9486 - val_loss: 0.1182 - val_fbeta_score: 0.7847 - val_accuracy: 0.9488
Epoch 25/25
253/253 [==============================] - 145s 571ms/step - loss: 0.1300 - fbeta_score: 0.7767 - accuracy: 0.9489 - val_loss: 0.1419 - val_fbeta_score: 0.7938 - val_accuracy: 0.9516

Modelo fazendo a classificação nos dados de teste. Seja paciente e aguarde...

Término do Fold  5  da Validação Cruzada.

Treinamento Concluído.
CPU times: user 5h 56min 52s, sys: 9min 14s, total: 6h 6min 6s
Wall time: 5h 10min 5s


# Saved model path
model_path_of_fold = os.path.join(PATH_MODELOS, 'pesos_fold_' + str(NUM_FOLD - 1) + '.h5')


# Create model
modelo_final = cria_modelo()


# Load saved model
modelo_final.load_weights(model_path_of_fold)


# Plot model performance
curva_aprendizado(modelo_cnn_satelite.history)


# The y_test item has all the predictions in the test data for each fold of the cross validation
len(y_teste)

5


# Create an array for results of one CV fold
resultado = np.array(y_teste[0])


# View fold results
resultado

array([[3.58537026e-02, 1.15087430e-03, 2.74071121e-03, ...,
        1.06197447e-02, 1.62826444e-03, 4.93933558e-02],
       [6.93802312e-02, 1.10839121e-03, 1.89726776e-03, ...,
        1.71465091e-02, 1.46847568e-03, 4.84707914e-02],
       [1.15033634e-01, 1.32205756e-03, 2.07197107e-03, ...,
        7.18102601e-05, 1.11096841e-03, 9.87202600e-02],
       ...,
       [1.28187209e-01, 3.08408635e-04, 5.84577210e-04, ...,
        1.44266465e-04, 6.78103126e-04, 9.32235867e-02],
       [9.32165861e-01, 1.80980866e-03, 2.02542841e-02, ...,
        5.01354458e-03, 1.23906862e-02, 1.58686966e-01],
       [5.99018559e-02, 6.51527080e-04, 3.68372235e-03, ...,
        6.84326887e-03, 1.73138711e-03, 9.66413543e-02]], dtype=float32)


# Convert array to dataframe and preview result
resultado_df = pd.DataFrame(resultado, columns = labels)
resultado_df.head()


# Shape
resultado_df.shape

(40479, 17)


# Loop to fill the array
for i in range(1, NFOLDS):
    resultado += np.array(y_teste[i])
    resultado /= NFOLDS


# Convert array to dataframe
resultado_df_final = pd.DataFrame(resultado, columns = labels)
resultado_df_final.head()


# Saving the predictions
resultado_df_final.to_csv(PATH_ARQUIVOS + 'resultado_previsoes.csv')


# Recording predictions and suggested classes in testing

# List for predictions
preds = []

# Loop to extract predictions
for i in range(resultado_df_final.shape[0]):
    a = resultado_df_final.iloc[[i]]
    a = a.apply(lambda x: x > THRES, axis = 1)
    a = a.transpose()
    a = a.loc[a[i] == True]
    ' '.join(list(a.index))
    preds.append(' '.join(list(a.index)))


# Recording the original labels and model predictions in csv format
# This is the file I'll submit to Kaggle
dados_teste_final['preds'] = preds
dados_teste_final.to_csv(PATH_ARQUIVOS + 'resultado_final.csv', index = False)


# Viewing 20 model predictions
dados_teste_final.head(20)

	image_name	tags
0	test_0	primary clear agriculture road water
1	test_1	primary clear agriculture road water
2	test_2	primary clear agriculture road water
3	test_3	primary clear agriculture road water
4	test_4	primary clear agriculture road water

	image_name	tags
0	test_0.jpg	[primary, clear, agriculture, road, water]
1	test_1.jpg	[primary, clear, agriculture, road, water]
2	test_2.jpg	[primary, clear, agriculture, road, water]
3	test_3.jpg	[primary, clear, agriculture, road, water]
4	test_4.jpg	[primary, clear, agriculture, road, water]

	image_name	tags
61186	file_9995.jpg	[primary, clear, agriculture, road, water]
61187	file_9996.jpg	[primary, clear, agriculture, road, water]
61188	file_9997.jpg	[primary, clear, agriculture, road, water]
61189	file_9998.jpg	[primary, clear, agriculture, road, water]
61190	file_9999.jpg	[primary, clear, agriculture, road, water]

	agriculture	artisinal_mine	bare_ground	blooming	blow_down	clear	cloudy	conventional_mine	cultivation	habitation	haze	partly_cloudy	primary	road	selective_logging	slash_burn	water
0	0.035854	0.001151	0.002741	3.312169e-02	0.004185	0.982465	0.000044	0.000278	0.023750	0.005646	0.000627	0.013735	0.999645	0.023195	0.010620	0.001628	0.049393
1	0.069380	0.001108	0.001897	2.924577e-02	0.007014	0.902134	0.000004	0.000149	0.065861	0.009277	0.000077	0.106229	0.999982	0.030755	0.017147	0.001468	0.048471
2	0.115034	0.001322	0.002072	1.965914e-05	0.000094	0.001064	0.016263	0.000399	0.030340	0.010436	0.000264	0.976215	0.983666	0.075785	0.000072	0.001111	0.098720
3	0.017205	0.000190	0.000708	1.400888e-02	0.000946	0.980363	0.000010	0.000054	0.012234	0.002051	0.000106	0.016535	0.999901	0.010341	0.003278	0.000325	0.023448
4	0.054526	0.000048	0.000135	9.120800e-08	0.000002	0.000796	0.420501	0.000011	0.005659	0.007255	0.000865	0.568739	0.532211	0.034072	0.000003	0.000026	0.078662

	agriculture	artisinal_mine	bare_ground	blooming	blow_down	clear	cloudy	conventional_mine	cultivation	habitation	haze	partly_cloudy	primary	road	selective_logging	slash_burn	water
0	0.017726	0.000241	0.001418	0.007395	0.002265	0.245403	0.000015	0.000047	0.013692	0.001906	0.000832	0.004331	0.251142	0.006566	0.002542	0.000602	0.021365
1	0.025142	0.000216	0.000907	0.015320	0.006049	0.243286	0.000001	0.000024	0.037599	0.003390	0.000075	0.010230	0.251197	0.006217	0.006709	0.001237	0.018200
2	0.049017	0.000511	0.001105	0.000022	0.000167	0.000125	0.000939	0.000391	0.034046	0.002138	0.000080	0.249998	0.249673	0.015629	0.000184	0.000667	0.039647
3	0.020377	0.000238	0.000840	0.009568	0.001577	0.188609	0.000002	0.000036	0.017641	0.001785	0.000045	0.081648	0.251196	0.008114	0.003946	0.000404	0.016709
4	0.031166	0.000176	0.000720	0.000010	0.000020	0.001138	0.109564	0.000060	0.009219	0.003895	0.001106	0.164217	0.146160	0.014621	0.000057	0.000089	0.025342

Satellite Image Classification¶

Project Definition¶

Dataset¶

Loading Packages¶

Checking Hardware Available on the Server¶

Loading Images¶

Preparing Data¶

Train Test Split¶

Choosing Image Resolution for Training the Model¶

Checking Label Distribution¶

Build Model¶

Train Model with Cross-Validation¶

Load the Saved Model¶

Training Results and Image Classification¶

Saving and Displaying Final Result¶

End¶

	image_name	tags
0	train_0	haze primary
1	train_1	agriculture clear primary water
2	train_2	clear primary
3	train_3	clear primary
4	train_4	agriculture clear habitation primary road

	image_name	tags
0	train_0.jpg	haze primary
1	train_1.jpg	agriculture clear primary water
2	train_2.jpg	clear primary
3	train_3.jpg	clear primary
4	train_4.jpg	agriculture clear habitation primary road