In [53]:

    

import os
import numpy as np
import pandas as pd
import scipy
import sklearn
import keras
from keras.models import Sequential
from keras.layers import Dense, Activation, Dropout, Convolution2D, Flatten, MaxPooling2D, Reshape, InputLayer
import cv2
from skimage import io
from keras.utils.np_utils import to_categorical

%matplotlib inline


    

In [54]:

    

seed = 410
rng = np.random.RandomState(seed)


    

In [55]:

    

rose_training_images = os.listdir("train/rose")
sunflower_training_images = os.listdir("train/sunflower")
training_rosepath = "train/rose/"
training_sfpath = "train/sunflower/"


    

In [56]:

    

# Data Preprocessing Method 1

training_images = []
training_label = []

# training data
for rose in rose_training_images:
    if rose.startswith('rose'):
        training_image = scipy.misc.imread(training_rosepath+rose)
        training_images.append(training_image)
        training_label.append(0) #for sunflower training images

for sunflower in sunflower_training_images:
    if sunflower.startswith('sunflower'):
        training_image = scipy.misc.imread(training_sfpath+sunflower)
        training_images.append(training_image)
        training_label.append(1) #for sunflower training images


    

In [58]:

    

for i in range(len(training_images)):
    training_images[i]=cv2.resize(training_images[i],(100,100))
      
training_images=np.array(training_images)
training_label=np.array(training_label)

training_label = to_categorical(training_label)


    

In [64]:

    

filters=25
filtersize1=(5,5)
filtersize2=(4,4)
epochs =7
batchsize=128
input_shape=(100,100,3)
hidden_num_units = 50


    

In [62]:

    

training_image_temp = training_images.reshape(-1, 100, 100, 3)   # requires 4 dimensions
## 3 is for RGB photo, if it's gray photo (1 color), set as 1


    

In [66]:

    

model = Sequential()

model.add(keras.layers.InputLayer(input_shape=input_shape))

model.add(keras.layers.convolutional.Conv2D(filters, filtersize1, strides=(1, 1), 
                                            padding='valid', data_format="channels_last", activation='relu'))
model.add(keras.layers.MaxPooling2D(pool_size=(2, 2)))
model.add(keras.layers.Flatten())

model.add(keras.layers.Dense(units=2, input_dim=55,activation='softmax'))   # 2 indicates 2 classes

model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])
model.fit(training_image_temp, training_label, epochs=epochs, batch_size=batchsize,validation_split=0.3)

model.summary()


    

    




Train on 77 samples, validate on 33 samples
Epoch 1/7
77/77 [==============================] - 1s - loss: 9.4806 - acc: 0.3506 - val_loss: 16.1181 - val_acc: 0.0000e+00
Epoch 2/7
77/77 [==============================] - 1s - loss: 4.6052 - acc: 0.7143 - val_loss: 16.1181 - val_acc: 0.0000e+00
Epoch 3/7
77/77 [==============================] - 1s - loss: 4.6052 - acc: 0.7143 - val_loss: 16.1181 - val_acc: 0.0000e+00
Epoch 4/7
77/77 [==============================] - 0s - loss: 4.6052 - acc: 0.7143 - val_loss: 16.1181 - val_acc: 0.0000e+00
Epoch 5/7
77/77 [==============================] - 0s - loss: 4.6052 - acc: 0.7143 - val_loss: 16.1181 - val_acc: 0.0000e+00
Epoch 6/7
77/77 [==============================] - 0s - loss: 4.6052 - acc: 0.7143 - val_loss: 16.1181 - val_acc: 0.0000e+00
Epoch 7/7
77/77 [==============================] - 0s - loss: 4.6052 - acc: 0.7143 - val_loss: 16.1181 - val_acc: 0.0000e+00
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
input_17 (InputLayer)        (None, 100, 100, 3)       0         
_________________________________________________________________
conv2d_33 (Conv2D)           (None, 96, 96, 25)        1900      
_________________________________________________________________
max_pooling2d_27 (MaxPooling (None, 48, 48, 25)        0         
_________________________________________________________________
flatten_15 (Flatten)         (None, 57600)             0         
_________________________________________________________________
dense_20 (Dense)             (None, 2)                 115202    
=================================================================
Total params: 117,102
Trainable params: 117,102
Non-trainable params: 0
_________________________________________________________________

In [104]:

    

# Predict with CNN trained model

testing_images_temp = testing_images.reshape(-1, 100, 100, 3)
pred = model.predict_classes(testing_images_temp)

testing_images = os.listdir("test/flower")
testing_path = "test/flower/"

img_name = rng.choice(testing_images)    # random choose a file name
testing_image_path = testing_path + img_name
testing_img = scipy.misc.imread(testing_image_path)  # without flatten keyword, we can show color photo below

test_index = int(img_name.split('.')[0])
if pred[test_index] == 0:
    print "Predicted as Rose"
else:
    print "Predicted as Sunflower"

pylab.imshow(testing_img)
pylab.axis('off')
pylab.show()


    

    




10/10 [==============================] - 0s
Predicted as Rose






    

In [133]:

    

img_name = rng.choice(testing_images)    # random choose a file name
testing_image_path = testing_path + img_name
testing_img = scipy.misc.imread(testing_image_path)  # without flatten keyword, we can show color photo below

test_index = int(img_name.split('.')[0])
if pred[test_index] == 0:
    print "Predicted as Rose"
else:
    print "Predicted as Sunflower"

pylab.imshow(testing_img)
pylab.axis('off')
pylab.show()

# Here, the training data only contains 55 images for rose and sunflower, not too much training data


    

    




Predicted as Rose






    

In [67]:

    

# Add more layers into CNN
## increase epochs
# validation accuracy doesn't improve

filters=25
filtersize1=(5,5)
filtersize2=(4,4)
epochs = 17
batchsize=128
input_shape=(100,100,3)
hidden_num_units = 50

model = Sequential()

model.add(keras.layers.InputLayer(input_shape=input_shape))

model.add(keras.layers.convolutional.Conv2D(filters, filtersize1, strides=(1, 1), 
                                            padding='valid', data_format="channels_last", activation='relu'))
model.add(keras.layers.MaxPooling2D(pool_size=(2, 2)))

model.add(keras.layers.convolutional.Conv2D(filters, filtersize1, strides=(1, 1), 
                                            padding='valid', data_format="channels_last", activation='relu'))
model.add(keras.layers.MaxPooling2D(pool_size=(2, 2)))

model.add(keras.layers.convolutional.Conv2D(filters, filtersize2, strides=(1, 1), 
                                            padding='valid', data_format="channels_last", activation='relu'))
model.add(keras.layers.MaxPooling2D(pool_size=(2, 2)))

model.add(keras.layers.Flatten())

model.add(keras.layers.Dense(units=hidden_num_units, activation='relu'))
model.add(keras.layers.Dense(units=2, input_dim=55,activation='softmax'))

model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])
model.fit(training_image_temp, training_label, epochs=epochs, batch_size=batchsize,validation_split=0.3)

model.summary()


    

    




Train on 77 samples, validate on 33 samples
Epoch 1/17
77/77 [==============================] - 2s - loss: 7.5357 - acc: 0.3377 - val_loss: 16.1181 - val_acc: 0.0000e+00
Epoch 2/17
77/77 [==============================] - 2s - loss: 4.6052 - acc: 0.7143 - val_loss: 16.1181 - val_acc: 0.0000e+00
Epoch 3/17
77/77 [==============================] - 2s - loss: 4.6052 - acc: 0.7143 - val_loss: 16.1181 - val_acc: 0.0000e+00
Epoch 4/17
77/77 [==============================] - 1s - loss: 4.6052 - acc: 0.7143 - val_loss: 16.1181 - val_acc: 0.0000e+00
Epoch 5/17
77/77 [==============================] - 1s - loss: 4.6052 - acc: 0.7143 - val_loss: 16.1181 - val_acc: 0.0000e+00
Epoch 6/17
77/77 [==============================] - 1s - loss: 4.6052 - acc: 0.7143 - val_loss: 16.1181 - val_acc: 0.0000e+00
Epoch 7/17
77/77 [==============================] - 1s - loss: 4.6052 - acc: 0.7143 - val_loss: 16.1181 - val_acc: 0.0000e+00
Epoch 8/17
77/77 [==============================] - 1s - loss: 4.6052 - acc: 0.7143 - val_loss: 16.1181 - val_acc: 0.0000e+00
Epoch 9/17
77/77 [==============================] - 1s - loss: 4.6052 - acc: 0.7143 - val_loss: 16.1181 - val_acc: 0.0000e+00
Epoch 10/17
77/77 [==============================] - 1s - loss: 4.6052 - acc: 0.7143 - val_loss: 16.1181 - val_acc: 0.0000e+00
Epoch 11/17
77/77 [==============================] - 1s - loss: 4.6052 - acc: 0.7143 - val_loss: 16.1181 - val_acc: 0.0000e+00
Epoch 12/17
77/77 [==============================] - 1s - loss: 4.6052 - acc: 0.7143 - val_loss: 16.1181 - val_acc: 0.0000e+00
Epoch 13/17
77/77 [==============================] - 1s - loss: 4.6052 - acc: 0.7143 - val_loss: 16.1181 - val_acc: 0.0000e+00
Epoch 14/17
77/77 [==============================] - 1s - loss: 4.6052 - acc: 0.7143 - val_loss: 16.1181 - val_acc: 0.0000e+00
Epoch 15/17
77/77 [==============================] - 2s - loss: 4.6052 - acc: 0.7143 - val_loss: 16.1181 - val_acc: 0.0000e+00
Epoch 16/17
77/77 [==============================] - 1s - loss: 4.6052 - acc: 0.7143 - val_loss: 16.1181 - val_acc: 0.0000e+00
Epoch 17/17
77/77 [==============================] - 2s - loss: 4.6052 - acc: 0.7143 - val_loss: 16.1181 - val_acc: 0.0000e+00
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
input_18 (InputLayer)        (None, 100, 100, 3)       0         
_________________________________________________________________
conv2d_34 (Conv2D)           (None, 96, 96, 25)        1900      
_________________________________________________________________
max_pooling2d_28 (MaxPooling (None, 48, 48, 25)        0         
_________________________________________________________________
conv2d_35 (Conv2D)           (None, 44, 44, 25)        15650     
_________________________________________________________________
max_pooling2d_29 (MaxPooling (None, 22, 22, 25)        0         
_________________________________________________________________
conv2d_36 (Conv2D)           (None, 19, 19, 25)        10025     
_________________________________________________________________
max_pooling2d_30 (MaxPooling (None, 9, 9, 25)          0         
_________________________________________________________________
flatten_16 (Flatten)         (None, 2025)              0         
_________________________________________________________________
dense_21 (Dense)             (None, 50)                101300    
_________________________________________________________________
dense_22 (Dense)             (None, 2)                 102       
=================================================================
Total params: 128,977
Trainable params: 128,977
Non-trainable params: 0
_________________________________________________________________