# Отчёт по лабораторной работе №3
## по теме: "Распознавание изображений"

---
Выполнили: Бригада 2, Мачулина Д.В., Бирюкова А.С., А-02-22 
---

## Задание 1
### 1. Создание блокнота и настройка среды

```python
from google.colab import drive
drive.mount('/content/drive')
import os
os.chdir('/content/drive/MyDrive/Colab Notebooks/is_lab3')

from tensorflow import keras
from tensorflow.keras import layers
from tensorflow.keras.models import Sequential
import matplotlib.pyplot as plt
import numpy as np
from sklearn.metrics import classification_report, confusion_matrix
from sklearn.metrics import ConfusionMatrixDisplay
```

---
### 2. Загрузка набора данных MNIST

```python
from keras.datasets import mnist
(X_train, y_train), (X_test, y_test) = mnist.load_data()
```

---
### 3. Разбиение набора данных на общучающие и тестовые (номер бригады - 2)
```python
from sklearn.model_selection import train_test_split

# объединяем в один набор
X = np.concatenate((X_train, X_test))
y = np.concatenate((y_train, y_test))

# разбиваем по вариантам
X_train, X_test, y_train, y_test = train_test_split(X, y,
                                                    test_size = 10000,
                                                    train_size = 60000,
                                                    random_state = 7)
# вывод размерностей
print('Shape of X train:', X_train.shape)
print('Shape of y train:', y_train.shape)
print('Shape of X test:', X_test.shape)
print('Shape of y test:', y_test.shape)
```
Shape of X train: (60000, 28, 28)
Shape of y train: (60000,)
Shape of X test: (10000, 28, 28)
Shape of y test: (10000,)

---
### 4. Предобработка данных

```python
# Зададим параметры данных и модели
num_classes = 10
input_shape = (28, 28, 1)

# Приведение входных данных к диапазону [0, 1]
X_train = X_train / 255
X_test = X_test / 255

# Расширяем размерность входных данных, чтобы каждое изображение имело
# размерность (высота, ширина, количество каналов)

X_train = np.expand_dims(X_train, -1)
X_test = np.expand_dims(X_test, -1)
print('Shape of transformed X train:', X_train.shape)
print('Shape of transformed X test:', X_test.shape)

# переведем метки в one-hot
y_train = keras.utils.to_categorical(y_train, num_classes)
y_test = keras.utils.to_categorical(y_test, num_classes)
print('Shape of transformed y train:', y_train.shape)
print('Shape of transformed y test:', y_test.shape)
```
Shape of transformed X train: (60000, 28, 28, 1)
Shape of transformed X test: (10000, 28, 28, 1)
Shape of transformed y train: (60000, 10)
Shape of transformed y test: (10000, 10)

---
### 5. Реализация и обучение модели свёрточной нейронной сети
```python
# создаем модель
model = Sequential()
model.add(layers.Conv2D(32, kernel_size=(3, 3), activation="relu", input_shape=input_shape))
model.add(layers.MaxPooling2D(pool_size=(2, 2)))
model.add(layers.Conv2D(64, kernel_size=(3, 3), activation="relu"))
model.add(layers.MaxPooling2D(pool_size=(2, 2)))
model.add(layers.Dropout(0.5))
model.add(layers.Flatten())
model.add(layers.Dense(num_classes, activation="softmax"))

model.summary()
```
<table>
    <thead>
        <tr>
            <th>Layer (type)</th>
            <th>Output Shape</th>
            <th>Param #</th>
        </tr>
    </thead>
    <tbody>
        <tr>
            <td>conv2d (Conv2D)</td>
            <td>(None, 26, 26, 32)</td>
            <td> 320 </td>
        </tr>
        <tr>
            <td>max_pooling2d (MaxPooling2D)</td>
            <td>(None, 13, 13, 32)  </td>
            <td>0</td>
        </tr>
        <tr>
            <td>conv2d_1 (Conv2D)   </td>
            <td>(None, 11, 11, 64) </td>
            <td>18,496</td>
        </tr>
        <tr>
            <td>max_pooling2d_1 (MaxPooling2D)</td>
            <td>(None, 5, 5, 64)</td>
            <td>0</td>
        </tr>
        <tr>
            <td>dropout (Dropout)</td>
            <td>(None, 5, 5, 64)</td>
            <td>0</td>
        </tr>
        <tr>
            <td>flatten (Flatten)  </td>
            <td>(None, 1600)</td>
            <td>0</td>
        </tr>
        <tr>
            <td>dense (Dense)</td>
            <td>(None, 10)</td>
            <td>16,010</td>
        </tr>
    </tbody>
</table>

Total params: 34,826 (136.04 KB)
Trainable params: 34,826 (136.04 KB)
Non-trainable params: 0 (0.00 B)

```python
# компилируем и обучаем модель
batch_size = 512
epochs = 15
model.compile(loss="categorical_crossentropy", optimizer="adam", metrics=["accuracy"])
model.fit(X_train, y_train, batch_size=batch_size, epochs=epochs, validation_split=0.1)
```

---
### 6. Оценка качества обучения на тестовых данных
```python
scores = model.evaluate(X_test, y_test)
print('Loss on test data:', scores[0])
print('Accuracy on test data:', scores[1])
```
Loss on test data: 0.04353996366262436
Accuracy on test data: 0.9876000285148621

---
 ### 7. Подача на вход обученной модели тестовых изображений
```python
# вывод тестового изображения и результата распознавания
n = 333
result = model.predict(X_test[n:n+1])
print('NN output:', result)
plt.imshow(X_test[n].reshape(28,28), cmap=plt.get_cmap('gray'))
plt.show()
print('Real mark: ', np.argmax(y_test[n]))
print('NN answer: ', np.argmax(result))
```

![](http://uit.mpei.ru/git/MachulinaDV/is_dnn/raw/branch/main/labworks/LW2/AE1_train_def.png)

Real mark:  3
NN answer:  3

```python
# вывод тестового изображения и результата распознавания
n = 222
result = model.predict(X_test[n:n+1])
print('NN output:', result)
plt.imshow(X_test[n].reshape(28,28), cmap=plt.get_cmap('gray'))
plt.show()
print('Real mark: ', np.argmax(y_test[n]))
print('NN answer: ', np.argmax(result))
```

![](http://uit.mpei.ru/git/MachulinaDV/is_dnn/raw/branch/main/labworks/LW2/AE2_train_def.png)

Real mark:  2
NN answer:  2

---
### 8. Вывод отчёта о качестве классификации тестовой выборки и матрицы ошибок для тестовой выборки
```python
# истинные метки классов
true_labels = np.argmax(y_test, axis=1)
# предсказанные метки классов
predicted_labels = np.argmax(model.predict(X_test), axis=1)
# отчет о качестве классификации
print(classification_report(true_labels, predicted_labels))
# вычисление матрицы ошибок
conf_matrix = confusion_matrix(true_labels, predicted_labels)
# отрисовка матрицы ошибок в виде "тепловой карты"
display = ConfusionMatrixDisplay(confusion_matrix=conf_matrix)
display.plot()
plt.show()
```

<table>
    <thead>
        <tr>
            <th>№(type)</th>
            <th>precision</th>
            <th>recall</th>
            <th>f1-score</th>
            <th>support</th>
        </tr>
    </thead>
    <tbody>
        <tr>
            <td>0</td>
            <td>0.99</td>
            <td>0.99</td>
            <td>0.99</td>
            <td>968</td>
        </tr>
        <tr>
            <td>1</td>
            <td>1.00</td>
            <td>0.99</td>
            <td>0.99</td>
            <td>1087</td>
        </tr>
        <tr>
            <td>2</td>
            <td>0.99</td>
            <td>0.99</td>
            <td>0.99</td>
            <td>1000</td>
        </tr>
        <tr>
            <td>3</td>
            <td>0.99</td>
            <td>0.98</td>
            <td>0.99</td>
            <td>1039</td>
        </tr>
        <tr>
            <td>4</td>
            <td>0.99</td>
            <td>0.99</td>
            <td>0.99</td>
            <td>966</td>
        </tr>
        <tr>
            <td>5</td>
            <td>0.98</td>
            <td>0.99</td>
            <td>0.99</td>
            <td>908</td>
        </tr>
        <tr>
            <td>6</td>
            <td>0.99</td>
            <td>0.99</td>
            <td>0.99</td>
            <td>972</td>
        </tr>
        <tr>
            <td>7</td>
            <td>0.98</td>
            <td>0.99</td>
            <td>0.98</td>
            <td>1060</td>
        </tr>
        <tr>
            <td>8</td>
            <td>0.98</td>
            <td>0.98</td>
            <td>0.98</td>
            <td>1015</td>
        </tr>
        <tr>
            <td>9</td>
            <td>0.98</td>
            <td>0.98</td>
            <td>0.98</td>
            <td>985</td>
        </tr>
        <tr>
        <td  colspan = 5></td>
        </tr>
        <tr>
            <td>accuracy</td>
            <td></td>
            <td></td>
            <td>0.99</td>
            <td>10000</td>
        </tr>
        <tr>
            <td>macro avg</td>
            <td>0.99</td>
            <td>0.99</td>
            <td>0.99</td>
            <td>10000</td>
        </tr>
        <tr>
            <td>weighted avg</td>
            <td>0.99</td>
            <td>0.99</td>
            <td>0.99</td>
            <td>10000</td>
        </tr>
    </tbody>
</table>


![]()
---
### 9. Загрузка, предобработка и подача собственных изображения
```python
# загрузка собственного изображения
from PIL import Image
file_data = Image.open('7.png')
file_data = file_data.convert('L') # перевод в градации серого
test_img = np.array(file_data)
# вывод собственного изображения
plt.imshow(test_img, cmap=plt.get_cmap('gray'))
plt.show()
# предобработка
test_img = test_img / 255
test_img = np.reshape(test_img, (1,28,28,1))
# распознавание
result = model.predict(test_img)
print('I think it\'s ', np.argmax(result))
```
![]()

```python
# загрузка собственного изображения
from PIL import Image
file_data = Image.open('5.png')
file_data = file_data.convert('L') # перевод в градации серого
test_img = np.array(file_data)
# вывод собственного изображения
plt.imshow(test_img, cmap=plt.get_cmap('gray'))
plt.show()
# предобработка
test_img = test_img / 255
test_img = np.reshape(test_img, (1,28,28,1))
# распознавание
result = model.predict(test_img)
print('I think it\'s ', np.argmax(result))
```
![]()

### 10. Загрузка модели из ЛР1. Оценка качества
```python
model = keras.models.load_model("best_model.keras")
model.summary()
```
<table>
    <thead>
        <tr>
            <th>Layer (type)</th>
            <th>Output Shape</th>
            <th>Param #</th>
        </tr>
    </thead>
    <tbody>
        <tr>
            <td>dense_4 (Dense)</td>
            <td>(None, 300)</td>
            <td>235,500</td>
        </tr>
        <tr>
            <td>dense_5 (Dense)</td>
            <td>(None, 10)</td>
            <td>3,010</td>
        </tr>
    </tbody>
</table>

 Total params: 238,512 (931.69 KB)
 Trainable params: 238,510 (931.68 KB)
 Non-trainable params: 0 (0.00 B)
 Optimizer params: 2 (12.00 B)


```python
# развернем каждое изображение 28*28 в вектор 784
X_train, X_test, y_train, y_test = train_test_split(X, y,
                                                    test_size = 10000,
                                                    train_size = 60000,
                                                    random_state = 7)
num_pixels = X_train.shape[1] * X_train.shape[2]
X_train = X_train.reshape(X_train.shape[0], num_pixels) / 255
X_test = X_test.reshape(X_test.shape[0], num_pixels) / 255
print('Shape of transformed X train:', X_train.shape)
print('Shape of transformed X train:', X_test.shape)

# переведем метки в one-hot
y_train = keras.utils.to_categorical(y_train, num_classes)
y_test = keras.utils.to_categorical(y_test, num_classes)
print('Shape of transformed y train:', y_train.shape)
print('Shape of transformed y test:', y_test.shape)
```
Shape of transformed X train: (60000, 784)
Shape of transformed X train: (10000, 784)
Shape of transformed y train: (60000, 10)
Shape of transformed y test: (10000, 10)

```python
# Оценка качества работы модели на тестовых данных
scores = model.evaluate(X_test, y_test)
print('Loss on test data:', scores[0])
print('Accuracy on test data:', scores[1])
```
Loss on test data: 0.37091827392578125
Accuracy on test data: 0.9013000130653381

---
### 11. Сравнение обученной модели сверточной сети и наилучшей модели полносвязной сети
<table>
    <thead>
        <tr>
            <th>Модель</th>
            <th>Количество настраиваемых параметров сети</th>
            <th>Количество эпох обучения</th>
            <th>Качество классификации тестовой выборки</th>
        </tr>
    </thead>
    <tbody>
        <tr>
            <th>Сверточная</th>
            <td align="center">34,826</td>
            <td align="center">15</td>
            <td align="center">0.9876</td>
        </tr>
        <tr>
            <th>Полносвязная</th>
            <td align="center">238,512</td>
            <td align="center">100</td>
            <td align="center">0.9013</td>
        </tr>
    </tbody>
</table>

Вывод:


## Задание 2
### В новом блокноте выполнили п.1-8 задания 1, изменив набор данных MNIST на CIFAR-10
### 1. Создание блокнота и настройка среды
```python
from google.colab import drive
drive.mount('/content/drive')
import os
os.chdir('/content/drive/MyDrive/Colab Notebooks/is_lab3')

from tensorflow import keras
from tensorflow.keras.models import Sequential
from tensorflow.keras import layers
import matplotlib.pyplot as plt
import numpy as np
from sklearn.metrics import classification_report, confusion_matrix
from sklearn.metrics import ConfusionMatrixDisplay
```

### 2.Загрузка набора данных и его разбиение на ообучащие и тестовые
```python
# загрузка датасета
from keras.datasets import cifar10
(X_train, y_train), (X_test, y_test) = cifar10.load_data()
```

```python
# создание своего разбиения датасета
from sklearn.model_selection import train_test_split


# объединяем в один набор
X = np.concatenate((X_train, X_test))
y = np.concatenate((y_train, y_test))

# разбиваем по вариантам
X_train, X_test, y_train, y_test = train_test_split(X, y,
                                                    test_size = 10000,
                                                    train_size = 50000,
                                                    random_state = 7)
# вывод размерностей
print('Shape of X train:', X_train.shape)
print('Shape of y train:', y_train.shape)
print('Shape of X test:', X_test.shape)
print('Shape of y test:', y_test.shape)
```
Shape of X train: (50000, 32, 32, 3)
Shape of y train: (50000, 1)
Shape of X test: (10000, 32, 32, 3)
Shape of y test: (10000, 1)

### 3. Вывод изображений с подписями классов
```python
class_names = ['airplane', 'automobile', 'bird', 'cat', 'deer',
 'dog', 'frog', 'horse', 'ship', 'truck']
plt.figure(figsize=(10,10))
for i in range(25):
 plt.subplot(5,5,i+1)
 plt.xticks([])
 plt.yticks([])
 plt.grid(False)
 plt.imshow(X_train[i])
 plt.xlabel(class_names[y_train[i][0]])
plt.show()
```
![]()

### 4. Предобработка данных
```python
# Зададим параметры данных и модели
num_classes = 10
input_shape = (32, 32, 3)

# Приведение входных данных к диапазону [0, 1]
X_train = X_train / 255
X_test = X_test / 255

print('Shape of transformed X train:', X_train.shape)
print('Shape of transformed X test:', X_test.shape)

# переведем метки в one-hot
y_train = keras.utils.to_categorical(y_train, num_classes)
y_test = keras.utils.to_categorical(y_test, num_classes)
print('Shape of transformed y train:', y_train.shape)
print('Shape of transformed y test:', y_test.shape)
```
Shape of transformed X train: (50000, 32, 32, 3)
Shape of transformed X test: (10000, 32, 32, 3)
Shape of transformed y train: (50000, 10)
Shape of transformed y test: (10000, 10)

---
### 5. Реализация и обучение модели свёрточной нейронной сети
```python
# создаем модель
model = Sequential()

# Блок 1
model.add(layers.Conv2D(32, (3, 3), padding="same",
                        activation="relu", input_shape=input_shape))
model.add(layers.BatchNormalization())
model.add(layers.Conv2D(32, (3, 3), padding="same", activation="relu"))
model.add(layers.BatchNormalization())
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Dropout(0.25))

# Блок 2
model.add(layers.Conv2D(64, (3, 3), padding="same", activation="relu"))
model.add(layers.BatchNormalization())
model.add(layers.Conv2D(64, (3, 3), padding="same", activation="relu"))
model.add(layers.BatchNormalization())
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Dropout(0.25))

model.add(layers.Flatten())
model.add(layers.Dense(128, activation='relu'))
model.add(layers.Dropout(0.5))
model.add(layers.Dense(num_classes, activation="softmax"))


model.summary()
```
<table>
    <thead>
        <tr>
            <th>Layer (type)</th>
            <th>Output Shape</th>
            <th>Param #</th>
        </tr>
    </thead>
    <tbody>
        <tr>
            <td>conv2d (Conv2D)</td>
            <td>(None, 32, 32, 32)</td>
            <td>896</td>
        </tr>
        <tr>
            <td>batch_normalization_6 (BatchNormalization)</td>
            <td>(None, 32, 32, 32)  </td>
            <td>128</td>
        </tr>
        <tr>
            <td>conv2d_13 (Conv2D)</td>
            <td>(None, 32, 32, 32)</td>
            <td>9,248</td>
        </tr>
        <tr>
            <td>batch_normalization_7 (BatchNormalization)</td>
            <td>(None, 32, 32, 32)</td>
            <td>128</td>
        </tr>
        <tr>
            <td>max_pooling2d_9 (MaxPooling2D)</td>
            <td>(None, 16, 16, 32)</td>
            <td>0</td>
        </tr>
        <tr>
            <td>dropout_6 (Dropout)</td>
            <td>(None, 16, 16, 32)</td>
            <td>0</td>
        </tr>
        <tr>
            <td>conv2d_14 (Conv2D)</td>
            <td>(None, 16, 16, 64)</td>
            <td>18,496</td>
        </tr>
        <tr>
            <td>batch_normalization_8 (BatchNormalization)</td>
            <td>(None, 16, 16, 64)</td>
            <td>256</td>
        </tr>
        <tr>
            <td>conv2d_15 (Conv2D)</td>
            <td>(None, 16, 16, 64)</td>
            <td>32,928</td>
        </tr>
        <tr>
            <td>batch_normalization_9 (BatchNormalization)</td>
            <td>(None, 16, 16, 64)</td>
            <td>256</td>
        </tr>
        <tr>
            <td>max_pooling2d_10 (MaxPooling2D)</td>
            <td>(None, 8, 8, 64)</td>
            <td>0</td>
        </tr>
        <tr>
            <td>dropout_7 (Dropout)</td>
            <td>(None, 8, 8, 64)</td>
            <td>0</td>
        </tr>
        <tr>
            <td>flatten_3 (Flatten) </td>
            <td>(None, 4096)</td>
            <td>0</td>
        </tr>
        <tr>
            <td>dense_6 (Dense)</td>
            <td>(None, 128)</td>
            <td>524,416</td>
        </tr>
        <tr>
            <td>dropout_8 (Dropout)</td>
            <td>(None, 128)</td>
            <td>0</td>
        </tr>
        <tr>
            <td>dense_7 (Dense)</td>
            <td>(None, 10)</td>
            <td>1,290</td>
        </tr>
    </tbody>
</table>
Total params: 592,042 (2.26 MB)
 Trainable params: 591,658 (2.26 MB)
 Non-trainable params: 384 (1.50 KB)

```python
batch_size = 64
epochs = 50
model.compile(loss="categorical_crossentropy", optimizer="adam", metrics=["accuracy"])
model.fit(X_train, y_train, batch_size=batch_size, epochs=epochs, validation_split=0.1)
```

### 6. Оценка качества обучения на тестовых данных
```python
scores = model.evaluate(X_test, y_test)
print('Loss on test data:', scores[0])
print('Accuracy on test data:', scores[1])
```

### 7. Подача на вход обученной модели тестовых изображений
```python
for n in [5,17]:
  result = model.predict(X_test[n:n+1])
  print('NN output:', result)

  plt.imshow(X_test[n].reshape(32,32,3), cmap=plt.get_cmap('gray'))
  plt.show()
  print('Real mark: ', np.argmax(y_test[n]))
  print('NN answer: ', np.argmax(result))
```
![]()
Real mark:  0
NN answer:  2

![]()
Real mark:  5
NN answer:  5

### 8. Вывод отчёта о качестве классификации тестовой выборки и матрицы ошибок для тестовой выборки
```python
# истинные метки классов
true_labels = np.argmax(y_test, axis=1)
# предсказанные метки классов
predicted_labels = np.argmax(model.predict(X_test), axis=1)

# отчет о качестве классификации
print(classification_report(true_labels, predicted_labels, target_names=class_names))
# вычисление матрицы ошибок
conf_matrix = confusion_matrix(true_labels, predicted_labels)
# отрисовка матрицы ошибок в виде "тепловой карты"
fig, ax = plt.subplots(figsize=(6, 6))
disp = ConfusionMatrixDisplay(confusion_matrix=conf_matrix,display_labels=class_names)
disp.plot(ax=ax, xticks_rotation=45)  # поворот подписей по X и приятная палитра
plt.tight_layout()  # чтобы всё влезло
plt.show()
```

<table>
    <thead>
        <tr>
            <th>class</th>
            <th>precision</th>
            <th>recall</th>
            <th>f1-score</th>
            <th>support</th>
        </tr>
    </thead>
    <tbody>
        <tr>
            <td>airplane</td>
            <td>0.85</td>
            <td>0.86</td>
            <td>0.86</td>
            <td>1013</td>
        </tr>
        <tr>
            <td>automobile</td>
            <td>0.93</td>
            <td>0.91</td>
            <td>0.92</td>
            <td>989</td>
        </tr>
        <tr>
            <td>bird</td>
            <td>0.75</td>
            <td>0.75</td>
            <td>0.75</td>
            <td>1018</td>
        </tr>
        <tr>
            <td>cat</td>
            <td>0.69</td>
            <td>0.66</td>
            <td>0.67</td>
            <td>1049</td>
        </tr>
        <tr>
            <td>deer</td>
            <td>0.79</td>
            <td>0.78</td>
            <td>0.78</td>
            <td>1009</td>
        </tr>
        <tr>
            <td>dog</td>
            <td>0.73</td>
            <td>0.68</td>
            <td>0.71</td>
            <td>978</td>
        </tr>
        <tr>
            <td>frog</td>
            <td>0.79</td>
            <td>0.90</td>
            <td>0.84</td>
            <td>981</td>
        </tr>
        <tr>
            <td>horse</td>
            <td>0.88</td>
            <td>0.84</td>
            <td>0.86</td>
            <td>986</td>
        </tr>
        <tr>
            <td>ship</td>
            <td>0.89</td>
            <td>0.92</td>
            <td>0.91</td>
            <td>1029</td>
        </tr>
        <tr>
            <td>truck</td>
            <td>0.88</td>
            <td>0.91</td>
            <td>0.89</td>
            <td>948</td>
        </tr>
        <tr>
        <td  colspan = 5></td>
        </tr>
        <tr>
            <td>accuracy</td>
            <td></td>
            <td></td>
            <td>0.82</td>
            <td>10000</td>
        </tr>
        <tr>
            <td>macro avg</td>
            <td>0.82</td>
            <td>0.82</td>
            <td>0.82</td>
            <td>10000</td>
        </tr>
        <tr>
            <td>weighted avg</td>
            <td>0.82</td>
            <td>0.82</td>
            <td>0.82</td>
            <td>10000</td>
        </tr>
    </tbody>
</table>

![]()

Вывод: