SW-Wiktor-Bombola/trees_nn.ipynb

import os
import numpy as np
import tensorflow as tf
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense, Dropout, Flatten, Activation, Conv2D, MaxPooling2D
from keras.preprocessing.image import ImageDataGenerator
from tensorflow.keras.optimizers import RMSprop
from tensorflow.keras.utils import plot_model
import matplotlib.pyplot as plt
import matplotlib.image as mpimg
import keras_tuner as kt
import keras

Konfiguracja

train_data_dir="./Trees"
validation_data_dir="./Trees"
batch_size=16
img_height, img_width = 60,80

Generowanie zdjęć

train_datagen = ImageDataGenerator(
    rescale=1./255, 
    shear_range=0.2, # view images from various angles
    zoom_range=0.2,
    horizontal_flip=True, # randomly flip img horizontally
    vertical_flip=True, # randomly flip img vertically
    validation_split=0.2) # set validation split

train_generator = train_datagen.flow_from_directory(
    train_data_dir,
    target_size=(img_height, img_width),
    batch_size=batch_size,
    class_mode='categorical',
    subset='training') # set as training data

validation_generator = train_datagen.flow_from_directory(
    validation_data_dir, # same directory as training data
    target_size=(img_height, img_width),
    batch_size=batch_size,
    class_mode='categorical',
    subset='validation')

Found 759 images belonging to 3 classes.
Found 187 images belonging to 3 classes.

def plot_image(fname, label, col):
    img = mpimg.imread(train_data_dir + '/' + fname)
    plt.subplot(1,3,col)
    plt.title(label)
    plt.imshow(img)
    
plt.figure(figsize=(10,5))
cat1_files = ['AS12_7/18.png','AS12_7/103.png','AS12_7/225.png']
for idx, f in enumerate(cat1_files):
    plot_image(f, f'Category 1: {idx+1}', idx + 1)
plt.show()

plt.figure(figsize=(10,5))
cat2_files = ['EA01a/15.png','EA01a/5.png','EA01a/12.png']
for idx, f in enumerate(cat2_files):
    plot_image(f, f'Category 2: {idx+1}', idx + 1)
plt.show()


plt.figure(figsize=(10,5))
cat3_files = ['EU43_6/135.png','EU43_6/195.png','EU43_6/16.png']
for idx, f in enumerate(cat3_files):
    plot_image(f, f'Category 2: {idx+1}', idx + 1)
plt.show()

CNN

model = Sequential()

model.add(Conv2D(32, (3,3), activation='relu', input_shape=(img_height, img_width, 3)))
model.add(MaxPooling2D((2,2)))

model.add(Conv2D(64, (3,3), activation='relu'))
model.add(MaxPooling2D((2,2)))

model.add(Conv2D(64, (3,3), activation='relu'))
model.add(MaxPooling2D((2,2)))

model.add(Flatten())

model.add(Dense(512, activation='relu'))
model.add(Dense(512, activation='relu'))

model.add(Dense(3, activation='softmax'))

plot_model(model, to_file='model1_plot.png', show_shapes=True,show_dtype=True, show_layer_names=True, expand_nested=True,)

print(model.summary())

Model: "sequential"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d (Conv2D)              (None, 58, 78, 32)        896       
_________________________________________________________________
max_pooling2d (MaxPooling2D) (None, 29, 39, 32)        0         
_________________________________________________________________
conv2d_1 (Conv2D)            (None, 27, 37, 64)        18496     
_________________________________________________________________
max_pooling2d_1 (MaxPooling2 (None, 13, 18, 64)        0         
_________________________________________________________________
conv2d_2 (Conv2D)            (None, 11, 16, 64)        36928     
_________________________________________________________________
max_pooling2d_2 (MaxPooling2 (None, 5, 8, 64)          0         
_________________________________________________________________
flatten (Flatten)            (None, 2560)              0         
_________________________________________________________________
dense (Dense)                (None, 512)               1311232   
_________________________________________________________________
dense_1 (Dense)              (None, 512)               262656    
_________________________________________________________________
dense_2 (Dense)              (None, 3)                 1539      
=================================================================
Total params: 1,631,747
Trainable params: 1,631,747
Non-trainable params: 0
_________________________________________________________________
None

model.compile(optimizer=RMSprop(learning_rate=0.0001),
             loss=tf.keras.losses.CategoricalCrossentropy(),
             metrics=['accuracy'])

history = model.fit(train_generator, epochs=16,  verbose=1,
                   validation_data = validation_generator, validation_steps = 4)

Epoch 1/16
48/48 [==============================] - 21s 419ms/step - loss: 0.8317 - accuracy: 0.6440 - val_loss: 0.7841 - val_accuracy: 0.8125
Epoch 2/16
48/48 [==============================] - 18s 376ms/step - loss: 0.6134 - accuracy: 0.7810 - val_loss: 0.3542 - val_accuracy: 0.8438
Epoch 3/16
48/48 [==============================] - 18s 377ms/step - loss: 0.4492 - accuracy: 0.8211 - val_loss: 0.2147 - val_accuracy: 0.9531
Epoch 4/16
48/48 [==============================] - 18s 367ms/step - loss: 0.2990 - accuracy: 0.9016 - val_loss: 0.4539 - val_accuracy: 0.8125
Epoch 5/16
48/48 [==============================] - 18s 379ms/step - loss: 0.2767 - accuracy: 0.8925 - val_loss: 0.0954 - val_accuracy: 0.9531
Epoch 6/16
48/48 [==============================] - 17s 364ms/step - loss: 0.2011 - accuracy: 0.9363 - val_loss: 0.1089 - val_accuracy: 0.9688
Epoch 7/16
48/48 [==============================] - 17s 362ms/step - loss: 0.2036 - accuracy: 0.9311 - val_loss: 0.0708 - val_accuracy: 1.0000
Epoch 8/16
48/48 [==============================] - 17s 361ms/step - loss: 0.1124 - accuracy: 0.9751 - val_loss: 0.5785 - val_accuracy: 0.7500
Epoch 9/16
48/48 [==============================] - 17s 360ms/step - loss: 0.1146 - accuracy: 0.9699 - val_loss: 0.0354 - val_accuracy: 1.0000
Epoch 10/16
48/48 [==============================] - 16s 341ms/step - loss: 0.1671 - accuracy: 0.9596 - val_loss: 0.0151 - val_accuracy: 1.0000
Epoch 11/16
48/48 [==============================] - 16s 327ms/step - loss: 0.0743 - accuracy: 0.9792 - val_loss: 0.0216 - val_accuracy: 1.0000
Epoch 12/16
48/48 [==============================] - 16s 332ms/step - loss: 0.0752 - accuracy: 0.9779 - val_loss: 0.0099 - val_accuracy: 1.0000
Epoch 13/16
48/48 [==============================] - 16s 328ms/step - loss: 0.1131 - accuracy: 0.9679 - val_loss: 0.0610 - val_accuracy: 0.9844
Epoch 14/16
48/48 [==============================] - 15s 318ms/step - loss: 0.1129 - accuracy: 0.9591 - val_loss: 0.0253 - val_accuracy: 1.0000
Epoch 15/16
48/48 [==============================] - 16s 329ms/step - loss: 0.0445 - accuracy: 0.9877 - val_loss: 0.0213 - val_accuracy: 1.0000
Epoch 16/16
48/48 [==============================] - 17s 346ms/step - loss: 0.0665 - accuracy: 0.9746 - val_loss: 0.0353 - val_accuracy: 0.9844

plt.plot(history.history['accuracy'], label='accuracy')
plt.plot(history.history['val_accuracy'], label = 'val_accuracy')
plt.xlabel('Epoch')
plt.ylabel('Accuracy')
plt.ylim([0.5, 1])
plt.legend(loc='lower right')

<matplotlib.legend.Legend at 0x17d175c4e20>

CNN + hyperparameter tuning

def build_model(hp):
    inputs = keras.Input(shape=(img_height, img_width, 3))
    x = inputs
    for i in range(hp.Int("cnn_layers", 1, 3)):
        x = Conv2D(
            hp.Int(f"filters_{i}", 32, 128, step=32),
            kernel_size=(3, 3),
            activation="relu",
        )(x)
        x = MaxPooling2D(pool_size=(2, 2))(x)
    x = Flatten()(x)

    # A hyperparamter for whether to use dropout layer.
    if hp.Boolean("dropout"):
        x = Dropout(0.5)(x)

    outputs = Dense(units=3, activation="softmax")(x)
    model = keras.Model(inputs=inputs, outputs=outputs)

    # Compile the model.
    model.compile(
        optimizer=RMSprop(learning_rate=0.0001),
             loss=tf.keras.losses.CategoricalCrossentropy(),
             metrics=['accuracy'],
    )
    return model

hp = kt.HyperParameters()
model = build_model(hp)

plot_model(model, to_file='model2_plot.png', show_shapes=True,show_dtype=True, show_layer_names=True, expand_nested=True,)

history = model.fit(train_generator, verbose=1, epochs=16,
                   validation_data = validation_generator, validation_steps = 4)

Epoch 1/16
48/48 [==============================] - 17s 342ms/step - loss: 0.7536 - accuracy: 0.6303 - val_loss: 0.5446 - val_accuracy: 0.9375
Epoch 2/16
48/48 [==============================] - 18s 369ms/step - loss: 0.4898 - accuracy: 0.8538 - val_loss: 0.3223 - val_accuracy: 0.9531
Epoch 3/16
48/48 [==============================] - 17s 362ms/step - loss: 0.3330 - accuracy: 0.8921 - val_loss: 0.1827 - val_accuracy: 0.9844
Epoch 4/16
48/48 [==============================] - 18s 380ms/step - loss: 0.2685 - accuracy: 0.9094 - val_loss: 0.2703 - val_accuracy: 0.9844
Epoch 5/16
48/48 [==============================] - 19s 403ms/step - loss: 0.2051 - accuracy: 0.9460 - val_loss: 0.2149 - val_accuracy: 0.9844
Epoch 6/16
48/48 [==============================] - 19s 403ms/step - loss: 0.1838 - accuracy: 0.9561 - val_loss: 0.1424 - val_accuracy: 0.9531
Epoch 7/16
48/48 [==============================] - 19s 394ms/step - loss: 0.1609 - accuracy: 0.9720 - val_loss: 0.1484 - val_accuracy: 0.9688
Epoch 8/16
48/48 [==============================] - 19s 404ms/step - loss: 0.1325 - accuracy: 0.9680 - val_loss: 0.0957 - val_accuracy: 0.9844
Epoch 9/16
48/48 [==============================] - 19s 406ms/step - loss: 0.1129 - accuracy: 0.9808 - val_loss: 0.1412 - val_accuracy: 0.9844
Epoch 10/16
48/48 [==============================] - 19s 401ms/step - loss: 0.1178 - accuracy: 0.9695 - val_loss: 0.0513 - val_accuracy: 1.0000
Epoch 11/16
48/48 [==============================] - 20s 408ms/step - loss: 0.0963 - accuracy: 0.9795 - val_loss: 0.0421 - val_accuracy: 1.0000
Epoch 12/16
48/48 [==============================] - 18s 387ms/step - loss: 0.0932 - accuracy: 0.9893 - val_loss: 0.0515 - val_accuracy: 1.0000
Epoch 13/16
48/48 [==============================] - 20s 416ms/step - loss: 0.0915 - accuracy: 0.9801 - val_loss: 0.1114 - val_accuracy: 0.9844
Epoch 14/16
48/48 [==============================] - 17s 356ms/step - loss: 0.0663 - accuracy: 0.9935 - val_loss: 0.0428 - val_accuracy: 1.0000
Epoch 15/16
48/48 [==============================] - 17s 360ms/step - loss: 0.0889 - accuracy: 0.9767 - val_loss: 0.0825 - val_accuracy: 0.9844
Epoch 16/16
48/48 [==============================] - 17s 347ms/step - loss: 0.0807 - accuracy: 0.9835 - val_loss: 0.0292 - val_accuracy: 1.0000

plt.plot(history.history['accuracy'], label='accuracy')
plt.plot(history.history['val_accuracy'], label = 'val_accuracy')
plt.xlabel('Epoch')
plt.ylabel('Accuracy')
plt.ylim([0.5, 1])
plt.legend(loc='lower right')

<matplotlib.legend.Legend at 0x17d1769efd0>

MLP

model = Sequential()
model.add(Flatten(input_shape=(img_height, img_width, 3), name="Input_layer"))
model.add(Dense(350, activation='relu'))
model.add(Dense(50, activation='relu'))
model.add(Dense(3, activation='softmax'))

plot_model(model, to_file='model3_plot.png', show_shapes=True,show_dtype=True, show_layer_names=True, expand_nested=True,)

model.compile(
        optimizer=RMSprop(learning_rate=0.0001),
             loss=tf.keras.losses.CategoricalCrossentropy(),
             metrics=['accuracy'],
    )

history = model.fit(train_generator, verbose=1, epochs=16,
                   validation_data = validation_generator, validation_steps = 4)

Epoch 1/16
48/48 [==============================] - 19s 388ms/step - loss: 1.2904 - accuracy: 0.5216 - val_loss: 0.6917 - val_accuracy: 0.6562
Epoch 2/16
48/48 [==============================] - 19s 404ms/step - loss: 0.6465 - accuracy: 0.7112 - val_loss: 0.7890 - val_accuracy: 0.6719
Epoch 3/16
48/48 [==============================] - 18s 383ms/step - loss: 0.5448 - accuracy: 0.7737 - val_loss: 0.4680 - val_accuracy: 0.8594
Epoch 4/16
48/48 [==============================] - 18s 368ms/step - loss: 0.4313 - accuracy: 0.8452 - val_loss: 0.5577 - val_accuracy: 0.7500
Epoch 5/16
48/48 [==============================] - 24s 501ms/step - loss: 0.4464 - accuracy: 0.8121 - val_loss: 0.2900 - val_accuracy: 0.7969
Epoch 6/16
48/48 [==============================] - 21s 438ms/step - loss: 0.3362 - accuracy: 0.8622 - val_loss: 0.2690 - val_accuracy: 0.7969
Epoch 7/16
48/48 [==============================] - 20s 426ms/step - loss: 0.3143 - accuracy: 0.8954 - val_loss: 0.3525 - val_accuracy: 0.8438
Epoch 8/16
48/48 [==============================] - 20s 426ms/step - loss: 0.2822 - accuracy: 0.9002 - val_loss: 0.2506 - val_accuracy: 0.9688
Epoch 9/16
48/48 [==============================] - 19s 398ms/step - loss: 0.2282 - accuracy: 0.9134 - val_loss: 0.3262 - val_accuracy: 0.9219
Epoch 10/16
48/48 [==============================] - 20s 409ms/step - loss: 0.1698 - accuracy: 0.9452 - val_loss: 0.3116 - val_accuracy: 0.9375
Epoch 11/16
48/48 [==============================] - 20s 409ms/step - loss: 0.2168 - accuracy: 0.9184 - val_loss: 0.2565 - val_accuracy: 0.9062
Epoch 12/16
48/48 [==============================] - 20s 409ms/step - loss: 0.2145 - accuracy: 0.9224 - val_loss: 0.1434 - val_accuracy: 1.0000
Epoch 13/16
48/48 [==============================] - 20s 425ms/step - loss: 0.2573 - accuracy: 0.9295 - val_loss: 0.1047 - val_accuracy: 0.9688
Epoch 14/16
48/48 [==============================] - 19s 405ms/step - loss: 0.1623 - accuracy: 0.9460 - val_loss: 0.1506 - val_accuracy: 0.9844
Epoch 15/16
48/48 [==============================] - 19s 395ms/step - loss: 0.1781 - accuracy: 0.9512 - val_loss: 0.0769 - val_accuracy: 0.9844
Epoch 16/16
48/48 [==============================] - 18s 383ms/step - loss: 0.2260 - accuracy: 0.9164 - val_loss: 0.2339 - val_accuracy: 0.9844

plt.plot(history.history['accuracy'], label='accuracy')
plt.plot(history.history['val_accuracy'], label = 'val_accuracy')
plt.xlabel('Epoch')
plt.ylabel('Accuracy')
plt.ylim([0.5, 1])
plt.legend(loc='lower right')

<matplotlib.legend.Legend at 0x17d195eb6d0>