（1） Keras handwritten numeral recognition and recognition of self written numbers

Training data is mnist, This is an example of the official website , I ran it over and saved the model parameters , Then use parameters to identify their own picture numbers .
The procedure consists of three main parts ：

1. Run the official website instance to save model parameters
2. Make your own pictures , Convert to the required format
3. Load model parameters , Test your photos

One 、 Model training and parameter saving

from __future__ import print_function
import keras
from keras.datasets import mnist
from keras.models import Sequential
from keras.layers import Dense, Dropout, Flatten
from keras.layers import Conv2D, MaxPooling2D
from keras import backend as K

batch_size = 128
num_classes = 10
epochs = 12

# input image dimensions
#  Enter the image dimension 
img_rows, img_cols = 28, 28

# the data, shuffled and split between train and test sets
#  Data sets for training and testing , After screening （ cleaning 、 The sequence of data samples is disordered ） And segmentation （ Split into training and test sets ）
(x_train, y_train), (x_test, y_test) = mnist.load_data()

#  Convert to the format required for input 
if K.image_data_format() == 'channels_first':  # Theano frame , Image channel in front 
    x_train = x_train.reshape(x_train.shape[0], 1, img_rows, img_cols)
    x_test = x_test.reshape(x_test.shape[0], 1, img_rows, img_cols)
    input_shape = (1, img_rows, img_cols)
else:  # TensorFlow frame , The image channel is in the back 
    x_train = x_train.reshape(x_train.shape[0], img_rows, img_cols, 1)
    x_test = x_test.reshape(x_test.shape[0], img_rows, img_cols, 1)
    input_shape = (img_rows, img_cols, 1)

#  Floating point numbers 
x_train = x_train.astype('float32')
x_test = x_test.astype('float32')
x_train /= 255
x_test /= 255
#  Number of training and testing 
print('x_train shape:', x_train.shape)
print(x_train.shape[0], 'train samples')
print(x_test.shape[0], 'test samples')

# convert class vectors to binary class matrices
#  The category vector is converted to 2 Classification matrix 
y_train = keras.utils.to_categorical(y_train, num_classes)
y_test = keras.utils.to_categorical(y_test, num_classes)

#  Build network 
model = Sequential()
model.add(Conv2D(32, kernel_size=(3, 3),
                 activation='relu',
                 input_shape=input_shape))
model.add(Conv2D(64, (3, 3), activation='relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.25))
model.add(Flatten())
model.add(Dense(128, activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(num_classes, activation='softmax'))

#  Loss function , Optimizer 
model.compile(loss=keras.losses.categorical_crossentropy,
              optimizer=keras.optimizers.Adadelta(),
              metrics=['accuracy'])
#  Start training 
model.fit(x_train, y_train,
          batch_size=batch_size,
          epochs=epochs,
          verbose=1,
          validation_data=(x_test, y_test))
#  test 
score = model.evaluate(x_test, y_test, verbose=0)
print('Test loss:', score[0])
print('Test accuracy:', score[1])

#  Save model parameters 
model.save('model.h5')

The training parameters are saved in the directory where the program is located model.h5 In the document

Two 、 Make your own pictures

The size is 2828 size , Black background , The font is white , It can be made in the drawing in the computer , First write the numbers with a black pen , Then reverse the color , Take a screenshot of each number , And then it's processed into 2828, Black and white photos .
I use Opencv Image processing done ：
What have I done 0-9 Named 10 A picture , In the same directory as the program , Of course, you can also deal with it in other ways

import cv2 as cv

for i in range(10):
    img = cv.imread("%d"%(i)+".png")
    img = cv.cvtColor(img, cv.COLOR_BGR2GRAY)

    res=cv.resize(img,(28,28),interpolation=cv.INTER_CUBIC)
    cv.imwrite("%d"%(i)+".png", res)
cv.waitKey(0)

3、 ... and 、 Identify your photos

Now load the model parameters , Then predict your own picture and output the predicted value

from __future__ import print_function
import matplotlib.pyplot as plt
import tensorflow as tf
import numpy as np

#  Import model 
model = tf.keras.models.load_model('./model.h5')

for i in range(10):
    test_my_img = plt.imread('%d'%(i)+'.png')
    img = test_my_img
    test_my_img = test_my_img.reshape(1,28,28,1)
    p = model.predict(test_my_img)
    print(np.argmax(p[0]))
    plt.imshow(img)
    plt.show()

The predicted values and input images are as follows ：

hold 9 Predicted success 7 了 , It is easy for the human eye to recognize , But it's a little difficult to distinguish the program
Recognize that there are many numbers on a large picture ：https://blog.csdn.net/X1996_/article/details/108889366