mnist数据集

训练模型

import tensorflow as tf

import keras
from keras.models import Sequential
from keras.layers import Dense,Dropout, Flatten,Conv2D, MaxPooling2D
# from keras.optimizers import SGD
from tensorflow.keras.optimizers import Adam,Nadam, SGD


from PIL import Image
import numpy as np
import matplotlib.pyplot as plt

print('tf',tf.__version__)
print('keras',keras.__version__)


# batch大小，每处理128个样本进行一次梯度更新
batch_size = 64
# 训练素材类别数
num_classes = 10
# 迭代次数
epochs = 5



f = np.load("mnist.npz")
x_train, y_train = f['x_train'], f['y_train']
x_test, y_test = f['x_test'], f['y_test']
f.close()
print(x_train.shape,"  ",y_train.shape)
print(x_test.shape,"  ",y_test.shape)

# im=plt.imshow(x_train[0],cmap="gray")
# plt.show()

## 维度合并784 = 28*28
x_train = x_train.reshape(60000, 784).astype('float32')
x_test = x_test.reshape(10000, 784).astype('float32')

## 归一化，像素点的值 转成 0-1 之间的数字
x_train /= 255
x_test /= 255

# print(x_train[0])

# 标签转换为独热码
y_train = tf.keras.utils.to_categorical(y_train, num_classes)
y_test = tf.keras.utils.to_categorical(y_test, num_classes)
# print(y_train[0]) ## 类似 [0. 0. 0. 0. 0. 1. 0. 0. 0. 0.]

print(x_train.shape,"  ",y_train.shape)
print(x_test.shape,"  ",y_test.shape)

# 构建模型
model = Sequential()
model.add(Dense(512, activation='relu',input_shape=(784,)))
model.add(Dense(256, activation='relu'))
model.add(Dense(num_classes, activation='softmax'))
model.summary()

# [编译模型] 配置模型，损失函数采用交叉熵，优化采用Adadelta，将识别准确率作为模型评估
model.compile(loss=tf.keras.losses.categorical_crossentropy, optimizer=tf.keras.optimizers.Adadelta(), metrics=['accuracy'])
#  validation_data为验证集
model.fit(x_train, y_train, batch_size=batch_size, epochs=epochs, verbose=1, validation_data=(x_test, y_test))

# 开始评估模型效果 # verbose=0为不输出日志信息
score = model.evaluate(x_test, y_test, verbose=0)
print('Test loss:', score[0])
print('Test accuracy:', score[1]) # 准确度


model.save('mnist_model_weights.h5') # 保存训练模型

训练结果

Epoch 4/5
938/938 [==============================] - 7s 7ms/step - loss: 1.5926 - accuracy: 0.7292 - val_loss: 1.4802 - val_accuracy: 0.7653
Epoch 5/5
938/938 [==============================] - 6s 6ms/step - loss: 1.4047 - accuracy: 0.7686 - val_loss: 1.2988 - val_accuracy: 0.7918
Test loss: 1.2988097667694092
Test accuracy: 0.7918000221252441

Process finished with exit code 0

测试模型

import tensorflow as tf

from PIL import Image
import numpy as np
from keras.models import load_model

# 构建模型
model = load_model('mnist_model_weights.h5') # 加载训练模型
# model.summary()

def read_image(img_name):
    im = Image.open(img_name).resize((28,28),Image.ANTIALIAS).convert('L') # 将要识别的图缩放到训练图一样的大小，并且灰度化
    data = np.array(im)
    return data

images=[]
images.append(read_image("test.png"))
# print(images)

X = np.array(images)
print(X.shape)
X=X.reshape(1, 784).astype('float32')
print(X.shape)
X /=255
# print(X[0:1])
result=model.predict(X[0:1])[0] # 识别出第一张图的结果，多张图的时候，把后面的[0] 去掉，返回的就是多张图结果
num=0 # 用来分析预测的结果
for i in range(len(result)): # result的长度是10
    # print(result[i]*255)
    if result[i]*255>result[num]*255: # 值越大，就越可能是结果
        num=i

print("预测结果",num)

将数据集转换为图片

#coding: utf-8
import os
import tensorflow as tf
import input_data
from PIL import Image

'''
函数功能：按照bmp格式提取mnist数据集中的图片
参数介绍:
    mnist_dir   mnist数据集存储的路径
    save_dir    提取结果存储的目录
'''

mint=tf.keras.datasets.mnist

def extract_mnist(mnist_dir, save_dir):
    rows = 28
    cols = 28

    # 加载mnist数据集
    # one_hot = True为默认打开"独热编码"
    mnist = input_data.read_data_sets(mnist_dir, one_hot=False)
    # 获取训练图片数量
    shape = mnist.train.images.shape
    images_train_count = shape[0]
    pixels_count_per_image = shape[1]
    # 获取训练标签数量=训练图片数量
    # 关闭"独热编码"后，labels的类型为[7 3 4 ... 5 6 8]
    labels = mnist.train.labels
    print(labels)
    exit(0)
    labels_train_count = labels.shape[0]

    if (images_train_count == labels_train_count):
        print("训练集共包含%d张图片，%d个标签" % (images_train_count, labels_train_count))
        print("每张图片包含%d个像素" % (pixels_count_per_image))
        print("数据类型为", mnist.train.images.dtype)

        # mnist图像数值的范围为[0,1], 需将其转换为[0,255]
        for current_image_id in range(images_train_count):
            for i in range(pixels_count_per_image):
                if mnist.train.images[current_image_id][i] != 0:
                    mnist.train.images[current_image_id][i] = 255

            if ((current_image_id + 1) % 50) == 0:
                print("已转换%d张，共需转换%d张" %
                      (current_image_id + 1, images_train_count))

        # 创建train images的保存目录, 按标签保存
        for i in range(10):
            dir = "%s/%s" % (save_dir, i)
            print(dir)
            if not os.path.exists(dir):
                os.mkdir(dir)

        # indices = [0, 0, 0, ..., 0]用来记录每个标签对应的图片数量
        indices = [0 for x in range(0, 10)]
        for i in range(images_train_count):
            new_image = Image.new("L", (cols, rows))
            # 遍历new_image 进行赋值
            for r in range(rows):
                for c in range(cols):
                    new_image.putpixel(
                        (r, c), int(mnist.train.images[i][c + r * cols]))

            # 获取第i张训练图片对应的标签
            label = labels[i]
            image_save_path = "%s/%s/%s.bmp" % (save_dir, label,
                                                indices[label])
            indices[label] += 1
            new_image.save(image_save_path)

            # 打印保存进度
            if ((i + 1) % 50) == 0:
                print("图片保存进度: 已保存%d张，共需保存%d张" % (i + 1, images_train_count))
    else:
        print("图片数量与标签数量不一致!")


if __name__ == '__main__':
    mnist_dir = "Mnist_Data"
    save_dir = "Mnist_Data_TrainImages"
    extract_mnist(mnist_dir, save_dir)

利用图片制作mnist格式数据集

import os
from PIL import Image
from array import *
from random import shuffle

# # 文件组织架构：
# ├──training-images
# │   └──0（类别为0的图像）
# │   ├──1（类别为1的图像）
# │   ├──2（类别为2的图像）
# │   ├──3（类别为3的图像）
# │   └──4（类别为4的图像）
# ├──test-images
# │   └──0（类别为0的图像）
# │   ├──1（类别为1的图像）
# │   ├──2（类别为2的图像）
# │   ├──3（类别为3的图像）
# │   └──4（类别为4的图像）
# └── mnist数据集制作.py（本脚本）

# Load from and save to
Names = [['./training-images', 'train'], ['./test-images', 'test']]

for name in Names:

    data_image = array('B')
    data_label = array('B')

    print(os.listdir(name[0]))
    FileList = []
    for dirname in os.listdir(name[0])[0:]:  # [1:] Excludes .DS_Store from Mac OS
        # print(dirname)
        path = os.path.join(name[0], dirname)
        # print(path)
        for filename in os.listdir(path):
            # print(filename)
            if filename.endswith(".png"):
                FileList.append(os.path.join(name[0] + '/', dirname + '/', filename))
        print(FileList)
    shuffle(FileList)  # Usefull for further segmenting the validation set

    for filename in FileList:

        label = int(filename.split('/')[2])
        print(filename)
        Im = Image.open(filename)
        # print(Im)

        pixel = Im.load()

        width, height = Im.size

        for x in range(0, width):
            for y in range(0, height):
                data_image.append(pixel[y, x])

        data_label.append(label)  # labels start (one unsigned byte each)

    hexval = "{0:#0{1}x}".format(len(FileList), 6)  # number of files in HEX

    # header for label array

    header = array('B')
    header.extend([0, 0, 8, 1, 0, 0])
    header.append(int('0x' + hexval[2:][:2], 16))
    header.append(int('0x' + hexval[2:][2:], 16))

    data_label = header + data_label

    # additional header for images array

    if max([width, height]) <= 256:
        header.extend([0, 0, 0, width, 0, 0, 0, height])
    else:
        raise ValueError('Image exceeds maximum size: 256x256 pixels');

    header[3] = 3  # Changing MSB for image data (0x00000803)

    data_image = header + data_image

    output_file = open(name[1] + '-images-idx3-ubyte', 'wb')
    data_image.tofile(output_file)
    output_file.close()

    output_file = open(name[1] + '-labels-idx1-ubyte', 'wb')
    data_label.tofile(output_file)
    output_file.close()

# 运行脚本得到四个文件test-images-idx3-ubyte、test-labels-idx1-ubyte、train-images-idx3-ubyte、train-labels-idx1-ubyte
# 在cmd中利用gzip -c train-labels-idx1-ubyte > train-labels-idx1-ubyte.gz命令对上述四个文件压缩得到最终的mnist格式数据集