charater-recognition/training.py at main · Nihith303/charater-recognition · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
import numpy as np
import pandas as pd
import np_utils
import matplotlib.pyplot as plt
import tensorflow as tf
from tensorflow import keras
from keras.models import Sequential
from keras.layers import Dense, Conv2D, MaxPooling2D, Flatten, Dropout, BatchNormalization
from keras.utils import to_categorical
from keras.preprocessing.image import ImageDataGenerator
from sklearn.model_selection import train_test_split

def load_data():
  # Load MNIST & NIST dataset
  digits_data_train = pd.read_csv("sample_data/mnist_train.csv")
  digits_data_test = pd.read_csv("sample_data/mnist_test.csv")
  letters_dataset = pd.read_csv("sample_data/A_Z Handwritten Data.csv")
  digits_data = pd.concat([digits_data_train, digits_data_test], ignore_index=True)
  # Rename correct label column name to label
  digits_data.rename(columns={'0':'label'}, inplace=True)
  letters_dataset.rename(columns={'0':'label'}, inplace=True)

  # Select 1000 samples from each class
  digits_data = digits_data.groupby('label').head(1000)
  letters_dataset = letters_dataset.groupby('label').head(1000)

  # Split data into X and Y for each type
  Y1 = digits_data['label']
  X1 = digits_data.drop('label', axis=1)
  Y2 = letters_dataset["label"]
  X2 = letters_dataset.drop("label", axis=1)

  # Split data into train and test set
  x_train1, x_test1, y_train1, y_test1 = train_test_split(X1, Y1, train_size=0.9)
  x_train2, x_test2, y_train2, y_test2 = train_test_split(X2, Y2, train_size=0.9)

  # Convert into numpy array to ease preprocessing
  x_train1 = x_train1.to_numpy()
  x_test1 = x_test1.to_numpy()
  x_train2 = x_train2.to_numpy()
  x_test2 = x_test2.to_numpy()

  # Add ten to match number of classes for one-hot
  y_train2 = y_train2 + 10
  y_test2 = y_test2 + 10

  # Convert Y into one-hot vectors
  y_train1 = to_categorical(y_train1, num_classes=36)
  y_test1 = to_categorical(y_test1, num_classes=36)
  y_train2 = to_categorical(y_train2, num_classes=36)
  y_test2 = to_categorical(y_test2, num_classes=36)

  # Reshape Xs into CNN input dimension
  x_train1 = x_train1.reshape(x_train1.shape[0], 28, 28, 1)
  x_test1 = x_test1.reshape(x_test1.shape[0], 28, 28, 1)
  x_train2 = x_train2.reshape(x_train2.shape[0], 28, 28, 1)
  x_test2 = x_test2.reshape(x_test2.shape[0], 28, 28, 1)

  # Combine each X and Y from each dataset
  x_train = np.concatenate((x_train1, x_train2), axis=0)
  x_test = np.concatenate((x_test1, x_test2), axis=0)
  y_train = np.concatenate((y_train1, y_train2), axis=0)
  y_test = np.concatenate((y_test1, y_test2), axis=0)

  return x_train, x_test, y_train, y_test

def preprocessing(x_train, x_test):
  x_train = x_train / 255
  x_test = x_test / 255
  return x_train, x_test


x_train, x_test, y_train, y_test = load_data()
(x_train, x_test) = preprocessing(x_train, x_test)
# Initialize Keras image generator
generator = ImageDataGenerator(rotation_range=10, zoom_range=0.1, width_shift_range=0.1, height_shift_range=0.1)

# CNN Model
model = Sequential([
    Conv2D(filters=32, kernel_size=3, input_shape=(28, 28, 1), activation="relu"),
    BatchNormalization(),
    Conv2D(filters=32, kernel_size=3, activation="relu"),
    BatchNormalization(),
    Conv2D(filters=32, kernel_size=5, strides=2, padding="same", activation="relu"),
    BatchNormalization(),
    Dropout(0.4),
    Conv2D(filters=64, kernel_size=3, activation="relu"),
    BatchNormalization(),
    Conv2D(filters=64, kernel_size=3, activation="relu"),
    BatchNormalization(),
    Conv2D(filters=64, kernel_size=5, strides=2, padding="same", activation="relu"),
    BatchNormalization(),
    Dropout(0.4),
    Conv2D(128, kernel_size=4, activation="relu"),
    BatchNormalization(),
    Flatten(),
    Dropout(0.4),
    Dense(36, activation="softmax")
])

model.compile(loss="categorical_crossentropy", optimizer="adam", metrics=["accuracy"])

history = model.fit_generator(generator.flow(x_train, y_train, batch_size=64), epochs=30, steps_per_epoch=x_train.shape[0]/64, validation_data=(x_test, y_test))