import os
from PIL import Image
import numpy as np
import seaborn as sb
import random
import shutil
from keras import backend as K
import matplotlib
matplotlib.use("Agg")
from sklearn.metrics import classification_report
from sklearn.metrics import confusion_matrix
import matplotlib.pyplot as plt
import sys
sys.modules['Image'] = Image
from keras.models import Sequential
from keras.layers import Dense, Dropout, Activation, Flatten
from keras.optimizers import Adam
from keras.layers.normalization import BatchNormalization
from keras.utils import np_utils
from keras.layers import Conv2D, MaxPooling2D, ZeroPadding2D, GlobalAveragePooling2D
from keras.preprocessing.image import ImageDataGenerator
from keras import regularizers
model = Sequential()
n_filters = 64
model.add(Conv2D(n_filters, (3, 3), input_shape=(50, 50, 3)))
model.add(MaxPooling2D((3, 3)))
model.add(Conv2D(n_filters, (3, 3)))
model.add(MaxPooling2D((3, 3)))
model.add(Flatten())
model.add(Dense(n_filters))
n_class = 43
model.add(Dense(n_class))
model.add(Activation('softmax'))
model.compile(loss='categorical_crossentropy',
optimizer='Adam',
metrics=['accuracy'])
model.build()
model.summary()
trainGen = ImageDataGenerator(rescale = 1/255,
rotation_range = 20,
zoom_range = 0.05,
width_shift_range = 0.1,
height_shift_range = 0.1,
shear_range = 0.05)
BS = 32
trainFlow = trainGen.flow_from_directory(directory = "test/train",
class_mode = "categorical",
target_size = (50, 50),
color_mode = "rgb",
shuffle = True,
batch_size = BS)
valueGen = ImageDataGenerator(rescale=1/255)
valueFlow = valueGen.flow_from_directory(directory = "test/val",
class_mode = "categorical",
target_size = (50, 50),
color_mode = "rgb",
shuffle = True,
batch_size = BS)
n_train = 27446
n_val = 3921
fit = model.fit_generator(trainFlow,
steps_per_epoch = n_train // BS,
epochs = 4,
validation_data = valueFlow,
validation_steps = n_val // BS)
acc = fit.history['acc']
valAcc = fit.history['val_acc']
loss = fit.history['loss']
valueLoss = fit.history['val_loss']
epochs = range(1, len(acc) + 1)
fig = plt.figure()
plt.plot(epochs, acc, 'b', label='Training accuracy')
plt.plot(epochs, valAcc, 'r', label='Validation accuracy')
plt.title('Training and validation accuracy')
plt.legend()
fig.savefig('trainAcc.png', dpi=fig.dpi)
fig = plt.figure()
plt.plot(epochs, loss, 'b', label='Training loss')
plt.plot(epochs, valueLoss, 'r', label='Validation loss')
plt.title('Training and validation loss')
plt.legend()
fig.savefig('trainLoss.png', dpi=fig.dpi)

dir_path = os.listdir('test/train')
for subdir in dir_path:
num = len(os.listdir('test/train/' + str(subdir)))
 print(str(subdir) + ":" + str(num))
number_files = len(dir_path)
print(number_files)

testGen = ImageDataGenerator(rescale=1/255)
testFlow = testGen.flow_from_directory(directory = "test/test",
class_mode = "categorical",
target_size = (50, 50),
color_mode = "rgb",
shuffle = False,
batch_size = BS)
n_test = 7842
pred_prob = model.predict_generator(testFlow, steps = (n_test // BS) + 1)
pred = np.argmax(pred_prob, axis=1)
print('Confusion Matrix')
confusionMatrix = confusion_matrix(testFlow.classes, pred)
print(confusionMatrix)
namesTarget = [str(i) for i in range(0,43)]
print(classification_report(testFlow.classes, pred, namesTarget=namesTarget))
result = sum(sum(confusionMatrix))
TP = [];
TN = [];
FN = [];
FP = [];
P = [];
N = [];
accuracy = [];
specificity = [];
sensitivity = [];
print("Class / Accuracyuracy / Sensitivityitivity / Specificityificity")
for i in np.arange(n_class):
TP.append(confusionMatrix[i, i])
P.append(sum(ConfusionMatrix[i,:]))
N.append(result - P[i])
FP.append(sum(confusionMatrix[:,i]) - TP[i])
FN.append(sum(confusionMatrix[i,:]) - TP[i])
TN.append(N[i] - FP[i])
accuracy_ = (TP[i] + TN[i])/result
accuracy.append(accuracy_)
specificity_ = TN[i] / (TN[i] + FP[i])
specificity.append(specificity_)
sensitivity_ = TP[i] / (TP[i] + FN[i])
sensitivity.append(sensitivity_)
 print(str(i) + ' / ' + str(accuracy_) + ' / ' + str(sensitivity_) + ' /
' + str(specificity_))
fig = plt.figure()
sb.heatmap(confusionMatrix)
fig.savefig('confusionMatrix.png', dpi=fig.dpi)