vgg19.py

# -*- coding: utf-8 -*-
"""vgg19.ipynb

Automatically generated by Colaboratory.

Original file is located at
    https://colab.research.google.com/drive/1gJ6aQTeeISXC_T92A-KX3DOdKHNghigq
"""

# Commented out IPython magic to ensure Python compatibility.
import os
import shutil
import cv2
import pandas as pd
import tensorflow as tf
from keras.applications.vgg19 import preprocess_input
import matplotlib.pyplot as plt
import tensorflow 
from keras.preprocessing.image import ImageDataGenerator
from keras.models import Sequential
from keras.layers import InputLayer, BatchNormalization, Dropout, Flatten, Dense, Activation, MaxPool2D, Conv2D
from keras.callbacks import EarlyStopping, ModelCheckpoint
from keras.applications.vgg19 import VGG19
from keras.preprocessing.image import load_img, img_to_array
from keras.metrics import Recall,Precision
# %matplotlib inline
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns
import os
from keras import models, layers
from keras.utils.np_utils import to_categorical
from keras.preprocessing.image import ImageDataGenerator
from sklearn.model_selection import train_test_split
from pathlib import Path
from sklearn.metrics import classification_report, confusion_matrix

from IPython.core.interactiveshell import InteractiveShell
InteractiveShell.ast_node_interactivity = 'all'

root_path = Path("/content/Fish_Dataset/Fish_Dataset")
sub_paths = root_path.glob(r'**/*.png')
#converting to list, will deal with generator later
sub_paths = list(sub_paths)

#creating temporary list to handle generator
data = []
for i in range(len(sub_paths)):
    #get the 2nd from last directory name
    label = os.path.split(os.path.split(sub_paths[i])[0])[1]
    data.append([str(sub_paths[i]), label])
    
#Appending data to dataframe
df = pd.DataFrame(data, columns=['path', 'label'])
#Removing ground truths 
df = df[df['label'].apply(lambda x: x[-2:] != 'GT')].reset_index(drop=True)

#Splitting into training, validation, and testing data
train_df, test_df = train_test_split(df, test_size=0.2, random_state=2, shuffle=True)
train_df.count()
test_df.count()

#Importing images using image data preprocessing provided from keras
train_generator = ImageDataGenerator(validation_split=0.2)
test_generator = ImageDataGenerator()

train_imgs = train_generator.flow_from_dataframe(
    dataframe = train_df,
    x_col = "path",
    y_col = "label",
    target_size = (224, 224),
    color_mode = "rgb",
    class_mode = "categorical",
    batch_size = 32,
    shuffle = True,
    subset = "training"
)

val_imgs = train_generator.flow_from_dataframe(
    dataframe = train_df,
    x_col = "path",
    y_col = "label",
    target_size = (224, 224),
    color_mode = "rgb",
    class_mode = "categorical",
    batch_size = 32,
    shuffle = True,
    subset = "validation"
)

test_imgs = test_generator.flow_from_dataframe(
    dataframe = test_df,
    x_col = "path",
    y_col = "label",
    target_size = (224, 224),
    color_mode = "rgb",
    class_mode = "categorical",
    batch_size = 32,
    shuffle = False
)

# Model Initialization

base_model = VGG19(input_shape=(224,224,3), 
                         include_top=False,
                         weights="imagenet")

# Freezing Layers

for layer in base_model.layers:
    layer.trainable=False

# Building Model
model=Sequential()
model.add(base_model)
model.add(Dropout(0.5))
model.add(Flatten())
model.add(BatchNormalization())
model.add(Dense(512,kernel_initializer='he_uniform'))
model.add(BatchNormalization())
model.add(Activation('relu'))
model.add(Dense(9,activation='softmax'))

# Summary

model.summary()

4

from keras.utils.vis_utils import plot_model
from IPython.display import SVG, Image
plot_model(model, to_file='model.png', show_shapes=True, show_layer_names=True)
Image('model.png',width=400, height=200)

# Model Compile 

OPT    = tensorflow.keras.optimizers.Adam(lr=0.001)

model.compile(loss='categorical_crossentropy',
              metrics=['accuracy',Precision(),Recall()],
              optimizer=OPT)

# Defining Callbacks

filepath = './best_weights.hdf5'

earlystopping = EarlyStopping(monitor = 'val_accuracy', 
                              mode = 'max' , 
                              patience = 5,
                              verbose = 1)

checkpoint    = ModelCheckpoint(filepath, 
                                monitor = 'val_accuracy', 
                                mode='max', 
                                save_best_only=True, 
                                verbose = 1)


callback_list = [earlystopping, checkpoint]

model_history=model.fit(train_imgs,
                        validation_data=val_imgs,
                        epochs = 20,
                        callbacks = callback_list,
                        verbose = 1)

import keras
model.save('./best_weights.hdf5')
#model = keras.models.load_model('./best_weights.hdf5')

plt.plot(model_history.history['accuracy'])
plt.plot(model_history.history['val_accuracy'])
plt.title('model accuracy')
plt.ylabel('accuracy')
plt.xlabel('epoch')
plt.legend(['train', 'validation'], loc='lower right')
plt.show()

plt.plot(model_history.history['loss'])
plt.plot(model_history.history['val_loss'])
plt.title('train set loss')

plt.ylabel('loss')
plt.xlabel('epoch')
plt.legend(['train', 'validation'], loc='upper right')
plt.show()

plt.plot(model_history.history['precision'])
plt.plot(model_history.history['val_precision'])
plt.title(' precision')
plt.ylabel('precision')
plt.xlabel('epoch')
plt.legend(['train', 'validation'], loc='upper right')
plt.show()


plt.plot(model_history.history['recall'])
plt.plot(model_history.history['val_recall'])
plt.title(' recall')
plt.ylabel('recall')
plt.xlabel('epoch')
plt.legend(['train', 'validation'], loc='upper right')
plt.show()

loss, accuracy , precision,recall=model.evaluate(test_imgs)
print('Test Accuracy: %.3f' % accuracy)
print('Test Precision: %.3f' % precision)
print('Test Recall: %.3f' % recall)
print('Test loss: %.3f' % loss)

y_pred = model.predict(test_imgs)
y_pred = np.argmax(y_pred,axis=1)

from sklearn.metrics import classification_report
print(classification_report(test_imgs.labels,y_pred))