TensorFlow+神经网络学习

---

一、学习目标

1. 熟悉神经网络框架的搭建
2. 完成对MNIST数据集的训练
3. 最终完成简易的手势识别项目

二、基本图像分类

1、使用 Fashion MNIST 数据集

该数据集包含 10 个类别的 70,000 个灰度图像。这些图像以低分辨率（28x28 像素）展示了单件衣物，如下所示：

2、搭建学习框架

import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import datasets, layers, models
import numpy as np

# 载入数据
fashion_mnist = keras.datasets.fashion_mnist
(train_images, train_labels), (test_images, test_labels) = fashion_mnist.load_data()
# 类名
class_names = ['T-shirt/top', 'Trouser', 'Pullover', 'Dress', 'Coat',
               'Sandal', 'Shirt', 'Sneaker', 'Bag', 'Ankle boot']
# 预处理数据
train_images = train_images / 255.0
test_images = test_images / 255.0
# 模型建立
model = keras.Sequential([
    keras.layers.Flatten(input_shape=(28, 28)),
    keras.layers.Dense(128, activation='relu'),
    keras.layers.Dense(10)
])
# 模型编译
model.compile(optimizer='adam',
              loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
              metrics=['accuracy'])
# 模型训练
model.fit(train_images, train_labels, epochs=10)
# 模型评估
test_loss, test_acc = model.evaluate(test_images,  test_labels, verbose=2)
print('\nTest accuracy:', test_acc)
# 模型预测
probability_model = tf.keras.Sequential([model, 
                                         tf.keras.layers.Softmax()])
predictions = probability_model.predict(test_images)
# 显示预测结果
for i in range(len(predictions)):
    label = np.argmax(predictions[i])
    print(class_names[int(label)])

三、卷积神经网络

1、使用 CIFAR10 数据集

CIFAR10 数据集包含 10 类，共 60000 张彩色图片，每类图片有 6000 张。此数据集中 50000 个样例被作为训练集，剩余 10000 个样例作为测试集。类之间相互度立，不存在重叠的部分，如下所示：

2、搭建学习框架

import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import datasets, layers, models

# 载入数据
(train_images, train_labels), (test_images, test_labels) = datasets.cifar10.load_data()
# 类名
class_names = ['airplane', 'automobile', 'bird', 'cat', 'deer',
               'dog', 'frog', 'horse', 'ship', 'truck']
# 预处理数据
train_images, test_images = train_images / 255.0, test_images / 255.0
# 模型建立
model = keras.Sequential([
    keras.layers.Conv2D(32, (3, 3), activation='relu', input_shape=(28, 28, 1)),
    keras.layers.MaxPooling2D((2, 2)),
    keras.layers.Conv2D(64, (3, 3), activation='relu'),
    keras.layers.MaxPooling2D((2, 2)),
    keras.layers.Conv2D(64, (3, 3), activation='relu'),

    keras.layers.Flatten(),
    keras.layers.Dense(64, activation='relu'),
    keras.layers.Dense(10)
])
# 模型编译
model.compile(optimizer='adam',
              loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
              metrics=['accuracy'])
# 模型训练
history = model.fit(train_images, train_labels, epochs=10, 
                    validation_data=(test_images, test_labels))
# 模型评估
test_loss, test_acc = model.evaluate(test_images,  test_labels, verbose=2)

四、利用卷积神经网络训练 Fashion MNIST 数据集

Fashion MNIST 数据集与 CIFAR10 数据集最重要的一个区别在于图像的维数不同， Fashion MNIST 数据集为灰度图像，而 CIFAR10 数据集为RGB彩色图像，若单纯调用卷积神经网络训练数据集会发生如下错误：

ValueError: Input 0 of layer sequential is incompatible with the layer: : expected min_ndim=4, found ndim=3. Full shape received: [32, 28, 28]

提示卷积神经网络输入数据最小维数min_ndim=4，所以在数据预处理环节还需进行如下操作：

train_images = train_images.reshape(train_images.shape[0], 28, 28, 1)
test_images = test_images.reshape(test_images.shape[0], 28, 28, 1)

最终代码如下：

import tensorflow as tf
from tensorflow import keras

# Helper libraries
import numpy as np

# 载入数据
fashion_mnist = keras.datasets.fashion_mnist
(train_images, train_labels), (test_images, test_labels) = fashion_mnist.load_data()
# 类名
class_names = ['T-shirt/top', 'Trouser', 'Pullover', 'Dress', 'Coat',
               'Sandal', 'Shirt', 'Sneaker', 'Bag', 'Ankle boot']
# 预处理数据
train_images = train_images.reshape(train_images.shape[0], 28, 28, 1)
test_images = test_images.reshape(test_images.shape[0], 28, 28, 1)

train_images = train_images / 255.0
test_images = test_images / 255.0
# 模型建立
model = keras.Sequential([
    keras.layers.Conv2D(32, (3, 3), activation='relu', input_shape=(28, 28, 1)),
    keras.layers.MaxPooling2D((2, 2)),
    keras.layers.Conv2D(64, (3, 3), activation='relu'),
    keras.layers.MaxPooling2D((2, 2)),
    keras.layers.Conv2D(128, (3, 3), activation='relu'),

    keras.layers.Flatten(),
    keras.layers.Dense(128, activation='relu'),
    keras.layers.Dense(10)
])
# 模型编译
model.compile(optimizer='adam',
              loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
              metrics=['accuracy'])
# 模型训练
model.fit(train_images, train_labels, epochs=10)
# 模型评估
test_loss, test_acc = model.evaluate(test_images,  test_labels, verbose=2)
print('\nTest accuracy:', test_acc)
# 模型预测
probability_model = tf.keras.Sequential([model,
                                         tf.keras.layers.Softmax()])
predictions = probability_model.predict(test_images)
# 显示预测结果
for i in range(len(predictions)):
    label = np.argmax(predictions[i])
    print(class_names[int(label)])