LeNet模型——tensorflow实现

版本:tensorflow2.0.0rcl
github地址 https://github.com/yang-ze-kang/LeNet5

MNIST数据集

MNIST数据集简介

1. 包含0~9手写数字
2. 60000个训练集、10000个测试集
3. 数据格式：28*28
4. 灰度图（单通道）

MNIST数据集加载

方法一：加载网络数据

mnist = tf.keras.datasets.mnist
(x_train, y_train), (x_test, y_test) = mnist.load_data()

方法二：将数据下载到本地电脑，加载本地电脑数据
数据下载地址：https://storage.googleapis.com/tensorflow/tf-keras-datasets/mnist.npz

f = np.load("path")
x_train, y_train = f['x_train'],f['y_train']
x_test, y_test = f['x_test'],f['y_test']
f.close()

导入成功检验

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

结果

MNIST数据集可视化

依赖第三方库

import matplotlib.pyplot as plt

显示图片及label

image_index = 123
print(y_train[image_index])     #查看随机一张图片的label
plt.imshow(x_train[image_index], cmap='Greys')  #图片显示
plt.show()

数据集格式转换

x_train = np.pad(x_train, ((0, 0), (2, 2), (2, 2)), 'constant', constant_values=0)  #将图片从28*28扩展成32*32
x_train = x_train.astype('float32')  #数据类型转换
x_train /= 255  #数据正则化
x_train = x_train.reshape(x_train.shape[0], 32, 32, 1)  #数据维度转换
print(x_train.shape)

tendorflow

模型类

Model

1. 实例化
   
2. 函数

sumary 查看模型
complie 编译时定义优化器、参数等
fit 训练模型
save 保存训练好的模型
evalute 评估模型

Sequetial

1. 实例化：只有一个参数layers
2. 继承自Model

卷积类Conv2D

filters, #卷积核个数
kernel_size, #卷积核大小
strides=(1,1), #步长
padding=’ ', #valid或SAME
data_format=None, #默认channels_last
activation, #激活函数

池化类AveragePooling2D

pool_size, #必须设置
stride
padding
data_format

LeNet模型

模型结构

模型构建

Model方法

class LeNet(tf.keras.Model):
    def __init__(self):
        super().__init__()
        #模型
        self.conv_layer_1 = tf.keras.layers.Conv2D(
            filters=6,
            kernel_size=(5, 5),
            padding='valid',
            activation=tf.nn.relu)

        self.pool_layer_1 = tf.keras.layers.MaxPool2D(
            pool_size=(2, 2),
            padding='same')

        self.conv_layer_2 = tf.keras.layers.Conv2D(
            filters=16,
            kernel_size=(5, 5),
            padding='valid',
            activation=tf.nn.relu)

        self.pool_layer_2 = tf.keras.layers.MaxPool2D(
            pool_size=(2, 2),
            padding='same')

        self.flatten = tf.keras.layers.Flatten()

        self.fc_layer_1 = tf.keras.layers.Dense(
            units=120,
            activation=tf.nn.relu)

        self.fc_layer_2 = tf.keras.layers.Dense(
            units=84,
            activation=tf.nn.relu)

        self.output_layer = tf.keras.layers.Dense(
            units=10,
            activation=tf.nn.relu)

    def call(self, inputs):
        x = self.conv_layer_1(inputs)
        x = self.pool_layer_1(x)
        x = self.conv_layer_2(x)
        x = self.pool_layer_2(x)
        x = self.flatten(x)
        x = self.fc_layer_1(x)
        x = self.fc_layer_2(x)
        output = self.output_layer(x)

        return output

Sequential方法

model = tf.keras.models.Sequential([
    tf.keras.layers.Conv2D(filters=6, kernel_size=(5, 5), padding='valid', activation=tf.nn.relu, input_shape=(32, 32, 1)),
    tf.keras.layers.MaxPool2D(pool_size=(2, 2), padding='same'),
    tf.keras.layers.Conv2D(filters=6, kernel_size=(5, 5), padding='valid', activation=tf.nn.relu),
    tf.keras.layers.MaxPool2D(pool_size=(2, 2), padding='same'),
    tf.keras.layers.Flatten(),
    tf.keras.layers.Dense(units=120, activation=tf.nn.relu),
    tf.keras.layers.Dense(units=84, activation=tf.nn.relu),
    tf.keras.layers.Dense(units=10, activation=tf.nn.relu)
])

模型summary结果

模型训练与预测

模型训练

超参数

epochs -> 数据被训练的次数
batch_size ->小批梯度下降每批的大小
learning_rate -> 学习率

优化：Adam

源码

#------------------------------【训练】---------------------------------

#超参数设置
num_epochs = 10
batch_size = 64
learning_rate = 0.01

#优化器
adam_optimizer = tf.keras.optimizers.Adam(learning_rate)

#编译
model.compile(optimizer=adam_optimizer,
               loss=tf.keras.losses.sparse_categorical_crossentropy,
               metrics=['accuracy'])

#训练
start_time = datetime.datetime.now()

model.fit(x=x_train,
         y=y_train,
         batch_size=batch_size,
         epochs=num_epochs)

endtime = datetime.datetime.now()

time_cost = endtime - start_time
print('time_cost = ', time_cost)

保存/加载模型

#保存/加载模型
model.save('lenet.h5')
#model = tf.keras.models.load_model('lenet.h5')

评估

#------------------------------【评估】---------------------------------

x_test = DataFormat(x_test)
print(x_test.shape)

print(model.evaluate(x_test, y_test))

预测

#------------------------------【预测】---------------------------------

image_index = 2333
print(x_test[image_index].shape)
plt.imshow(x_test[image_index].reshape(32, 32), cmap='Greys')
plt.show()

pred = model.predict(x_test[image_index].reshape(1, 32, 32, 1))
print(pred.argmax())