一、构建CNN模型

from keras import layersfrom keras import modelsmodel = models.Sequential()model.add(layers.Conv2D(32, (3, 3),           activation='relu',           input_shape=(28, 28, 1))          )model.add(layers.MaxPooling2D((2, 2)))model.add(layers.Conv2D(64, (3, 3), activation='relu'))model.add(layers.MaxPooling2D((2, 2)))model.add(layers.Conv2D(64, (3, 3), activation='relu'))model.add(layers.Flatten())model.add(layers.Dropout(0.25))model.add(layers.Dense(64, activation='relu'))model.add(layers.Dense(10, activation='softmax'))

卷积神经网络接收形状为(image_height, image_width, image_channels)的输入张量（不包括批量维度）。本例中设置卷积神经网络处理大小为(28, 28, 1) 的输入张量，这正是MNIST 图像的格式。我们向第一层传入参数input_shape=(28, 28, 1) 来完成此设置。我们来看一下目前卷积神经网络的架构。

model.summary()Model: "sequential"_________________________________________________________________Layer (type)                 Output Shape              Param #   =================================================================conv2d (Conv2D)              (None, 26, 26, 32)        320       _________________________________________________________________max_pooling2d (MaxPooling2D) (None, 13, 13, 32)        0         _________________________________________________________________conv2d_1 (Conv2D)            (None, 11, 11, 64)        18496     _________________________________________________________________max_pooling2d_1 (MaxPooling2 (None, 5, 5, 64)          0         _________________________________________________________________conv2d_2 (Conv2D)            (None, 3, 3, 64)          36928     _________________________________________________________________flatten (Flatten)            (None, 576)               0         _________________________________________________________________dropout (Dropout)            (None, 576)               0         _________________________________________________________________dense (Dense)                (None, 64)                36928     _________________________________________________________________dense_1 (Dense)              (None, 10)                650       =================================================================Total params: 93,322Trainable params: 93,322Non-trainable params: 0_________________________________________________________________

二、图片下载与查看

from keras.datasets import mnistfrom keras.utils    import to_categoricalfrom keras          import datasets(train_images,train_labels), (test_images,test_labels)  = mnist.load_data()dir(datasets)#可以看看内置的所有数据集

查看图片

import matplotlib.pyplot as pltplt.imshow(train_images[200] , cmap=plt.cm.binary)plt.show()

三、模型训练&准确率评估

train_images = train_images.reshape((60000, 28, 28, 1))train_images = train_images.astype('float32') / 255train_labels = to_categorical(train_labels)

test_images  = test_images.reshape((10000, 28, 28, 1))test_images  = test_images.astype('float32') / 255test_labels  = to_categorical(test_labels)

model.compile(optimizer='rmsprop',              loss='categorical_crossentropy',              metrics=['accuracy'])model.fit(train_images, train_labels, epochs=5, batch_size=64)

test_loss, test_acc = model.evaluate(test_images, test_labels)test_acc 0.9914000034332275

#看看预测的到底准不准呢

y_pred = model.predict(test_images)import numpy as nppred = np.argmax(y_pred, axis=1)import matplotlib.pyplot as plt#看看第2990个数字是啥，我们预测的是8，看看图片也是8，挺准的steps = 2990print('pred: ',pred[steps])pred:  8plt.imshow(test_images[steps] , cmap=plt.cm.binary)plt.show()

四、预测自己的手写数字

模型训练好了，准确率挺高，但是实际有没有用，还需要用自己的数据进行测试，打开自己在画图板里面随便写几个数字，然后单个截图保存后进行预测。

分别截图后保存成img3、img4、img5......，下面进行预处理，处理成和模型训练一样的数据才能预测。

from keras.preprocessing.image import load_img,img_to_arrayimport matplotlib.pyplot as pltimport numpy as nppath = 'C:/Users/伍正祥/Desktop/img5.jpg'#读取图片、调整图片大小，转换成灰度 help(load_img)img = load_img(path, target_size=(28, 28),color_mode="grayscale")#255-为了调成白底，系统灰度转换自动给处理成黑底了img = 255-img_to_array(img)#查看自己加载的图片plt.imshow(img , cmap=plt.cm.binary)plt.show()#进行数据预测img = img.astype('float32')/255img = img.reshape((1, 28, 28, 1))y_pred = model.predict(img)print('预测数字：',np.argmax(y_pred, axis=1)[0]